ODIn/spark-instrumented-optimizer
1
0
Fork 0
Code Issues Pull requests Projects Releases Wiki Activity
spark-instrumented-optimizer/sql/core/benchmarks/DateTimeBenchmark-jdk11-results.txt

458 lines
37 KiB
Plaintext
Raw Normal View History

[SPARK-31527][SQL][TESTS][FOLLOWUP] Add a benchmark test for datetime add/subtract interval operations ### What changes were proposed in this pull request? With https://github.com/apache/spark/pull/28310, the operation of date +/- interval(m, d, 0) has been improved a lot. According to the benchmark results, about 75% time cost is reduced because of no casting date to timestamp back and forth. In this PR, we add a benchmark for these operations, and timestamp +/- interval operations as accessories. ### Why are the changes needed? Performance test coverage, since these operations are missing in the DateTimeBenchmark. ### Does this PR introduce any user-facing change? No, just test ### How was this patch tested? regenerated benchmark results Closes #28369 from yaooqinn/SPARK-31527-F. Authored-by: Kent Yao <yaooqinn@hotmail.com> Signed-off-by: Wenchen Fan <wenchen@databricks.com>
2020-04-28 11:39:28 -04:00
================================================================================================
datetime +/- interval
================================================================================================
[SPARK-31527][SQL][TESTS][FOLLOWUP] Fix the number of rows in `DateTimeBenchmark` ### What changes were proposed in this pull request? - Changed to the number of rows in benchmark cases from 3 to the actual number `N`. - Regenerated benchmark results in the environment: | Item | Description | | ---- | ----| | Region | us-west-2 (Oregon) | | Instance | r3.xlarge | | AMI | ubuntu/images/hvm-ssd/ubuntu-bionic-18.04-amd64-server-20190722.1 (ami-06f2f779464715dc5) | | Java | OpenJDK 64-Bit Server VM 1.8.0_242 and OpenJDK 64-Bit Server VM 11.0.6+10 | ### Why are the changes needed? The changes are needed to have: - Correct benchmark results - Base line for other perf improvements that can be checked in the same environment. ### Does this PR introduce _any_ user-facing change? No ### How was this patch tested? By running the benchmark and checking its output. Closes #28440 from MaxGekk/SPARK-31527-DateTimeBenchmark-followup. Authored-by: Max Gekk <max.gekk@gmail.com> Signed-off-by: Takeshi Yamamuro <yamamuro@apache.org>
2020-05-03 20:39:50 -04:00
OpenJDK 64-Bit Server VM 11.0.7+10-post-Ubuntu-2ubuntu218.04 on Linux 4.15.0-1063-aws
Intel(R) Xeon(R) CPU E5-2670 v2 @ 2.50GHz
[SPARK-31527][SQL][TESTS][FOLLOWUP] Add a benchmark test for datetime add/subtract interval operations ### What changes were proposed in this pull request? With https://github.com/apache/spark/pull/28310, the operation of date +/- interval(m, d, 0) has been improved a lot. According to the benchmark results, about 75% time cost is reduced because of no casting date to timestamp back and forth. In this PR, we add a benchmark for these operations, and timestamp +/- interval operations as accessories. ### Why are the changes needed? Performance test coverage, since these operations are missing in the DateTimeBenchmark. ### Does this PR introduce any user-facing change? No, just test ### How was this patch tested? regenerated benchmark results Closes #28369 from yaooqinn/SPARK-31527-F. Authored-by: Kent Yao <yaooqinn@hotmail.com> Signed-off-by: Wenchen Fan <wenchen@databricks.com>
2020-04-28 11:39:28 -04:00
datetime +/- interval: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative
------------------------------------------------------------------------------------------------------------------------
[SPARK-31630][SQL] Fix perf regression by skipping timestamps rebasing after some threshold ### What changes were proposed in this pull request? Skip timestamps rebasing after a global threshold when there is no difference between Julian and Gregorian calendars. This allows to avoid checking hash maps of switch points, and fixes perf regressions in `toJavaTimestamp()` and `fromJavaTimestamp()`. ### Why are the changes needed? The changes fix perf regressions of conversions to/from external type `java.sql.Timestamp`. Before (see the PR's results https://github.com/apache/spark/pull/28440): ``` ================================================================================================ Conversion from/to external types ================================================================================================ OpenJDK 64-Bit Server VM 1.8.0_252-8u252-b09-1~18.04-b09 on Linux 4.15.0-1063-aws Intel(R) Xeon(R) CPU E5-2670 v2 2.50GHz To/from Java's date-time: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative ------------------------------------------------------------------------------------------------------------------------ From java.sql.Timestamp 376 388 10 13.3 75.2 1.1X Collect java.sql.Timestamp 1878 1937 64 2.7 375.6 0.2X ``` After: ``` ================================================================================================ Conversion from/to external types ================================================================================================ OpenJDK 64-Bit Server VM 1.8.0_252-8u252-b09-1~18.04-b09 on Linux 4.15.0-1063-aws Intel(R) Xeon(R) CPU E5-2670 v2 2.50GHz To/from Java's date-time: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative ------------------------------------------------------------------------------------------------------------------------ From java.sql.Timestamp 249 264 24 20.1 49.8 1.7X Collect java.sql.Timestamp 1503 1523 24 3.3 300.5 0.3X ``` Perf improvements in average of: 1. From java.sql.Timestamp is ~ 34% 2. To java.sql.Timestamps is ~16% ### Does this PR introduce _any_ user-facing change? No ### How was this patch tested? By existing test suites `DateTimeUtilsSuite` and `RebaseDateTimeSuite`. Closes #28441 from MaxGekk/opt-rebase-common-threshold. Authored-by: Max Gekk <max.gekk@gmail.com> Signed-off-by: Wenchen Fan <wenchen@databricks.com>
2020-05-05 10:11:53 -04:00
date + interval(m) 1496 1569 104 6.7 149.6 1.0X
date + interval(m, d) 1514 1526 17 6.6 151.4 1.0X
date + interval(m, d, ms) 6231 6253 30 1.6 623.1 0.2X
date - interval(m) 1481 1487 9 6.8 148.1 1.0X
date - interval(m, d) 1550 1552 2 6.5 155.0 1.0X
date - interval(m, d, ms) 6269 6272 4 1.6 626.9 0.2X
timestamp + interval(m) 3017 3056 54 3.3 301.7 0.5X
timestamp + interval(m, d) 3146 3148 3 3.2 314.6 0.5X
timestamp + interval(m, d, ms) 3446 3460 20 2.9 344.6 0.4X
timestamp - interval(m) 3045 3059 19 3.3 304.5 0.5X
timestamp - interval(m, d) 3147 3164 25 3.2 314.7 0.5X
timestamp - interval(m, d, ms) 3425 3442 25 2.9 342.5 0.4X
[SPARK-31527][SQL][TESTS][FOLLOWUP] Add a benchmark test for datetime add/subtract interval operations ### What changes were proposed in this pull request? With https://github.com/apache/spark/pull/28310, the operation of date +/- interval(m, d, 0) has been improved a lot. According to the benchmark results, about 75% time cost is reduced because of no casting date to timestamp back and forth. In this PR, we add a benchmark for these operations, and timestamp +/- interval operations as accessories. ### Why are the changes needed? Performance test coverage, since these operations are missing in the DateTimeBenchmark. ### Does this PR introduce any user-facing change? No, just test ### How was this patch tested? regenerated benchmark results Closes #28369 from yaooqinn/SPARK-31527-F. Authored-by: Kent Yao <yaooqinn@hotmail.com> Signed-off-by: Wenchen Fan <wenchen@databricks.com>
2020-04-28 11:39:28 -04:00
[SPARK-29320][TESTS] Compare `sql/core` module in JDK8/11 (Part 1) ### What changes were proposed in this pull request? This PR regenerates the `sql/core` benchmarks in JDK8/11 to compare the result. In general, we compare the ratio instead of the time. However, in this PR, the average time is compared. This PR should be considered as a rough comparison. **A. EXPECTED CASES(JDK11 is faster in general)** - [x] BloomFilterBenchmark (JDK11 is faster except one case) - [x] BuiltInDataSourceWriteBenchmark (JDK11 is faster at CSV/ORC) - [x] CSVBenchmark (JDK11 is faster except five cases) - [x] ColumnarBatchBenchmark (JDK11 is faster at `boolean`/`string` and some cases in `int`/`array`) - [x] DatasetBenchmark (JDK11 is faster with `string`, but is slower for `long` type) - [x] ExternalAppendOnlyUnsafeRowArrayBenchmark (JDK11 is faster except two cases) - [x] ExtractBenchmark (JDK11 is faster except HOUR/MINUTE/SECOND/MILLISECONDS/MICROSECONDS) - [x] HashedRelationMetricsBenchmark (JDK11 is faster) - [x] JSONBenchmark (JDK11 is much faster except eight cases) - [x] JoinBenchmark (JDK11 is faster except five cases) - [x] OrcNestedSchemaPruningBenchmark (JDK11 is faster in nine cases) - [x] PrimitiveArrayBenchmark (JDK11 is faster) - [x] SortBenchmark (JDK11 is faster except `Arrays.sort` case) - [x] UDFBenchmark (N/A, values are too small) - [x] UnsafeArrayDataBenchmark (JDK11 is faster except one case) - [x] WideTableBenchmark (JDK11 is faster except two cases) **B. CASES WE NEED TO INVESTIGATE MORE LATER** - [x] AggregateBenchmark (JDK11 is slower in general) - [x] CompressionSchemeBenchmark (JDK11 is slower in general except `string`) - [x] DataSourceReadBenchmark (JDK11 is slower in general) - [x] DateTimeBenchmark (JDK11 is slightly slower in general except `parsing`) - [x] MakeDateTimeBenchmark (JDK11 is slower except two cases) - [x] MiscBenchmark (JDK11 is slower except ten cases) - [x] OrcV2NestedSchemaPruningBenchmark (JDK11 is slower) - [x] ParquetNestedSchemaPruningBenchmark (JDK11 is slower except six cases) - [x] RangeBenchmark (JDK11 is slower except one case) `FilterPushdownBenchmark/InExpressionBenchmark/WideSchemaBenchmark` will be compared later because it took long timer. ### Why are the changes needed? According to the result, there are some difference between JDK8/JDK11. This will be a baseline for the future improvement and comparison. Also, as a reproducible environment, the following environment is used. - Instance: `r3.xlarge` - OS: `CentOS Linux release 7.5.1804 (Core)` - JDK: - `OpenJDK Runtime Environment (build 1.8.0_222-b10)` - `OpenJDK Runtime Environment 18.9 (build 11.0.4+11-LTS)` ### Does this PR introduce any user-facing change? No. ### How was this patch tested? This is a test-only PR. We need to run benchmark. Closes #26003 from dongjoon-hyun/SPARK-29320. Authored-by: Dongjoon Hyun <dhyun@apple.com> Signed-off-by: Dongjoon Hyun <dhyun@apple.com>
2019-10-03 11:58:25 -04:00
================================================================================================
Extract components
================================================================================================
[SPARK-31527][SQL][TESTS][FOLLOWUP] Fix the number of rows in `DateTimeBenchmark` ### What changes were proposed in this pull request? - Changed to the number of rows in benchmark cases from 3 to the actual number `N`. - Regenerated benchmark results in the environment: | Item | Description | | ---- | ----| | Region | us-west-2 (Oregon) | | Instance | r3.xlarge | | AMI | ubuntu/images/hvm-ssd/ubuntu-bionic-18.04-amd64-server-20190722.1 (ami-06f2f779464715dc5) | | Java | OpenJDK 64-Bit Server VM 1.8.0_242 and OpenJDK 64-Bit Server VM 11.0.6+10 | ### Why are the changes needed? The changes are needed to have: - Correct benchmark results - Base line for other perf improvements that can be checked in the same environment. ### Does this PR introduce _any_ user-facing change? No ### How was this patch tested? By running the benchmark and checking its output. Closes #28440 from MaxGekk/SPARK-31527-DateTimeBenchmark-followup. Authored-by: Max Gekk <max.gekk@gmail.com> Signed-off-by: Takeshi Yamamuro <yamamuro@apache.org>
2020-05-03 20:39:50 -04:00
OpenJDK 64-Bit Server VM 11.0.7+10-post-Ubuntu-2ubuntu218.04 on Linux 4.15.0-1063-aws
Intel(R) Xeon(R) CPU E5-2670 v2 @ 2.50GHz
[SPARK-29320][TESTS] Compare `sql/core` module in JDK8/11 (Part 1) ### What changes were proposed in this pull request? This PR regenerates the `sql/core` benchmarks in JDK8/11 to compare the result. In general, we compare the ratio instead of the time. However, in this PR, the average time is compared. This PR should be considered as a rough comparison. **A. EXPECTED CASES(JDK11 is faster in general)** - [x] BloomFilterBenchmark (JDK11 is faster except one case) - [x] BuiltInDataSourceWriteBenchmark (JDK11 is faster at CSV/ORC) - [x] CSVBenchmark (JDK11 is faster except five cases) - [x] ColumnarBatchBenchmark (JDK11 is faster at `boolean`/`string` and some cases in `int`/`array`) - [x] DatasetBenchmark (JDK11 is faster with `string`, but is slower for `long` type) - [x] ExternalAppendOnlyUnsafeRowArrayBenchmark (JDK11 is faster except two cases) - [x] ExtractBenchmark (JDK11 is faster except HOUR/MINUTE/SECOND/MILLISECONDS/MICROSECONDS) - [x] HashedRelationMetricsBenchmark (JDK11 is faster) - [x] JSONBenchmark (JDK11 is much faster except eight cases) - [x] JoinBenchmark (JDK11 is faster except five cases) - [x] OrcNestedSchemaPruningBenchmark (JDK11 is faster in nine cases) - [x] PrimitiveArrayBenchmark (JDK11 is faster) - [x] SortBenchmark (JDK11 is faster except `Arrays.sort` case) - [x] UDFBenchmark (N/A, values are too small) - [x] UnsafeArrayDataBenchmark (JDK11 is faster except one case) - [x] WideTableBenchmark (JDK11 is faster except two cases) **B. CASES WE NEED TO INVESTIGATE MORE LATER** - [x] AggregateBenchmark (JDK11 is slower in general) - [x] CompressionSchemeBenchmark (JDK11 is slower in general except `string`) - [x] DataSourceReadBenchmark (JDK11 is slower in general) - [x] DateTimeBenchmark (JDK11 is slightly slower in general except `parsing`) - [x] MakeDateTimeBenchmark (JDK11 is slower except two cases) - [x] MiscBenchmark (JDK11 is slower except ten cases) - [x] OrcV2NestedSchemaPruningBenchmark (JDK11 is slower) - [x] ParquetNestedSchemaPruningBenchmark (JDK11 is slower except six cases) - [x] RangeBenchmark (JDK11 is slower except one case) `FilterPushdownBenchmark/InExpressionBenchmark/WideSchemaBenchmark` will be compared later because it took long timer. ### Why are the changes needed? According to the result, there are some difference between JDK8/JDK11. This will be a baseline for the future improvement and comparison. Also, as a reproducible environment, the following environment is used. - Instance: `r3.xlarge` - OS: `CentOS Linux release 7.5.1804 (Core)` - JDK: - `OpenJDK Runtime Environment (build 1.8.0_222-b10)` - `OpenJDK Runtime Environment 18.9 (build 11.0.4+11-LTS)` ### Does this PR introduce any user-facing change? No. ### How was this patch tested? This is a test-only PR. We need to run benchmark. Closes #26003 from dongjoon-hyun/SPARK-29320. Authored-by: Dongjoon Hyun <dhyun@apple.com> Signed-off-by: Dongjoon Hyun <dhyun@apple.com>
2019-10-03 11:58:25 -04:00
cast to timestamp: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative
------------------------------------------------------------------------------------------------------------------------
[SPARK-31630][SQL] Fix perf regression by skipping timestamps rebasing after some threshold ### What changes were proposed in this pull request? Skip timestamps rebasing after a global threshold when there is no difference between Julian and Gregorian calendars. This allows to avoid checking hash maps of switch points, and fixes perf regressions in `toJavaTimestamp()` and `fromJavaTimestamp()`. ### Why are the changes needed? The changes fix perf regressions of conversions to/from external type `java.sql.Timestamp`. Before (see the PR's results https://github.com/apache/spark/pull/28440): ``` ================================================================================================ Conversion from/to external types ================================================================================================ OpenJDK 64-Bit Server VM 1.8.0_252-8u252-b09-1~18.04-b09 on Linux 4.15.0-1063-aws Intel(R) Xeon(R) CPU E5-2670 v2 2.50GHz To/from Java's date-time: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative ------------------------------------------------------------------------------------------------------------------------ From java.sql.Timestamp 376 388 10 13.3 75.2 1.1X Collect java.sql.Timestamp 1878 1937 64 2.7 375.6 0.2X ``` After: ``` ================================================================================================ Conversion from/to external types ================================================================================================ OpenJDK 64-Bit Server VM 1.8.0_252-8u252-b09-1~18.04-b09 on Linux 4.15.0-1063-aws Intel(R) Xeon(R) CPU E5-2670 v2 2.50GHz To/from Java's date-time: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative ------------------------------------------------------------------------------------------------------------------------ From java.sql.Timestamp 249 264 24 20.1 49.8 1.7X Collect java.sql.Timestamp 1503 1523 24 3.3 300.5 0.3X ``` Perf improvements in average of: 1. From java.sql.Timestamp is ~ 34% 2. To java.sql.Timestamps is ~16% ### Does this PR introduce _any_ user-facing change? No ### How was this patch tested? By existing test suites `DateTimeUtilsSuite` and `RebaseDateTimeSuite`. Closes #28441 from MaxGekk/opt-rebase-common-threshold. Authored-by: Max Gekk <max.gekk@gmail.com> Signed-off-by: Wenchen Fan <wenchen@databricks.com>
2020-05-05 10:11:53 -04:00
cast to timestamp wholestage off 332 336 5 30.1 33.2 1.0X
cast to timestamp wholestage on 333 344 10 30.0 33.3 1.0X
[SPARK-29320][TESTS] Compare `sql/core` module in JDK8/11 (Part 1) ### What changes were proposed in this pull request? This PR regenerates the `sql/core` benchmarks in JDK8/11 to compare the result. In general, we compare the ratio instead of the time. However, in this PR, the average time is compared. This PR should be considered as a rough comparison. **A. EXPECTED CASES(JDK11 is faster in general)** - [x] BloomFilterBenchmark (JDK11 is faster except one case) - [x] BuiltInDataSourceWriteBenchmark (JDK11 is faster at CSV/ORC) - [x] CSVBenchmark (JDK11 is faster except five cases) - [x] ColumnarBatchBenchmark (JDK11 is faster at `boolean`/`string` and some cases in `int`/`array`) - [x] DatasetBenchmark (JDK11 is faster with `string`, but is slower for `long` type) - [x] ExternalAppendOnlyUnsafeRowArrayBenchmark (JDK11 is faster except two cases) - [x] ExtractBenchmark (JDK11 is faster except HOUR/MINUTE/SECOND/MILLISECONDS/MICROSECONDS) - [x] HashedRelationMetricsBenchmark (JDK11 is faster) - [x] JSONBenchmark (JDK11 is much faster except eight cases) - [x] JoinBenchmark (JDK11 is faster except five cases) - [x] OrcNestedSchemaPruningBenchmark (JDK11 is faster in nine cases) - [x] PrimitiveArrayBenchmark (JDK11 is faster) - [x] SortBenchmark (JDK11 is faster except `Arrays.sort` case) - [x] UDFBenchmark (N/A, values are too small) - [x] UnsafeArrayDataBenchmark (JDK11 is faster except one case) - [x] WideTableBenchmark (JDK11 is faster except two cases) **B. CASES WE NEED TO INVESTIGATE MORE LATER** - [x] AggregateBenchmark (JDK11 is slower in general) - [x] CompressionSchemeBenchmark (JDK11 is slower in general except `string`) - [x] DataSourceReadBenchmark (JDK11 is slower in general) - [x] DateTimeBenchmark (JDK11 is slightly slower in general except `parsing`) - [x] MakeDateTimeBenchmark (JDK11 is slower except two cases) - [x] MiscBenchmark (JDK11 is slower except ten cases) - [x] OrcV2NestedSchemaPruningBenchmark (JDK11 is slower) - [x] ParquetNestedSchemaPruningBenchmark (JDK11 is slower except six cases) - [x] RangeBenchmark (JDK11 is slower except one case) `FilterPushdownBenchmark/InExpressionBenchmark/WideSchemaBenchmark` will be compared later because it took long timer. ### Why are the changes needed? According to the result, there are some difference between JDK8/JDK11. This will be a baseline for the future improvement and comparison. Also, as a reproducible environment, the following environment is used. - Instance: `r3.xlarge` - OS: `CentOS Linux release 7.5.1804 (Core)` - JDK: - `OpenJDK Runtime Environment (build 1.8.0_222-b10)` - `OpenJDK Runtime Environment 18.9 (build 11.0.4+11-LTS)` ### Does this PR introduce any user-facing change? No. ### How was this patch tested? This is a test-only PR. We need to run benchmark. Closes #26003 from dongjoon-hyun/SPARK-29320. Authored-by: Dongjoon Hyun <dhyun@apple.com> Signed-off-by: Dongjoon Hyun <dhyun@apple.com>
2019-10-03 11:58:25 -04:00
[SPARK-31527][SQL][TESTS][FOLLOWUP] Fix the number of rows in `DateTimeBenchmark` ### What changes were proposed in this pull request? - Changed to the number of rows in benchmark cases from 3 to the actual number `N`. - Regenerated benchmark results in the environment: | Item | Description | | ---- | ----| | Region | us-west-2 (Oregon) | | Instance | r3.xlarge | | AMI | ubuntu/images/hvm-ssd/ubuntu-bionic-18.04-amd64-server-20190722.1 (ami-06f2f779464715dc5) | | Java | OpenJDK 64-Bit Server VM 1.8.0_242 and OpenJDK 64-Bit Server VM 11.0.6+10 | ### Why are the changes needed? The changes are needed to have: - Correct benchmark results - Base line for other perf improvements that can be checked in the same environment. ### Does this PR introduce _any_ user-facing change? No ### How was this patch tested? By running the benchmark and checking its output. Closes #28440 from MaxGekk/SPARK-31527-DateTimeBenchmark-followup. Authored-by: Max Gekk <max.gekk@gmail.com> Signed-off-by: Takeshi Yamamuro <yamamuro@apache.org>
2020-05-03 20:39:50 -04:00
OpenJDK 64-Bit Server VM 11.0.7+10-post-Ubuntu-2ubuntu218.04 on Linux 4.15.0-1063-aws
Intel(R) Xeon(R) CPU E5-2670 v2 @ 2.50GHz
[SPARK-29320][TESTS] Compare `sql/core` module in JDK8/11 (Part 1) ### What changes were proposed in this pull request? This PR regenerates the `sql/core` benchmarks in JDK8/11 to compare the result. In general, we compare the ratio instead of the time. However, in this PR, the average time is compared. This PR should be considered as a rough comparison. **A. EXPECTED CASES(JDK11 is faster in general)** - [x] BloomFilterBenchmark (JDK11 is faster except one case) - [x] BuiltInDataSourceWriteBenchmark (JDK11 is faster at CSV/ORC) - [x] CSVBenchmark (JDK11 is faster except five cases) - [x] ColumnarBatchBenchmark (JDK11 is faster at `boolean`/`string` and some cases in `int`/`array`) - [x] DatasetBenchmark (JDK11 is faster with `string`, but is slower for `long` type) - [x] ExternalAppendOnlyUnsafeRowArrayBenchmark (JDK11 is faster except two cases) - [x] ExtractBenchmark (JDK11 is faster except HOUR/MINUTE/SECOND/MILLISECONDS/MICROSECONDS) - [x] HashedRelationMetricsBenchmark (JDK11 is faster) - [x] JSONBenchmark (JDK11 is much faster except eight cases) - [x] JoinBenchmark (JDK11 is faster except five cases) - [x] OrcNestedSchemaPruningBenchmark (JDK11 is faster in nine cases) - [x] PrimitiveArrayBenchmark (JDK11 is faster) - [x] SortBenchmark (JDK11 is faster except `Arrays.sort` case) - [x] UDFBenchmark (N/A, values are too small) - [x] UnsafeArrayDataBenchmark (JDK11 is faster except one case) - [x] WideTableBenchmark (JDK11 is faster except two cases) **B. CASES WE NEED TO INVESTIGATE MORE LATER** - [x] AggregateBenchmark (JDK11 is slower in general) - [x] CompressionSchemeBenchmark (JDK11 is slower in general except `string`) - [x] DataSourceReadBenchmark (JDK11 is slower in general) - [x] DateTimeBenchmark (JDK11 is slightly slower in general except `parsing`) - [x] MakeDateTimeBenchmark (JDK11 is slower except two cases) - [x] MiscBenchmark (JDK11 is slower except ten cases) - [x] OrcV2NestedSchemaPruningBenchmark (JDK11 is slower) - [x] ParquetNestedSchemaPruningBenchmark (JDK11 is slower except six cases) - [x] RangeBenchmark (JDK11 is slower except one case) `FilterPushdownBenchmark/InExpressionBenchmark/WideSchemaBenchmark` will be compared later because it took long timer. ### Why are the changes needed? According to the result, there are some difference between JDK8/JDK11. This will be a baseline for the future improvement and comparison. Also, as a reproducible environment, the following environment is used. - Instance: `r3.xlarge` - OS: `CentOS Linux release 7.5.1804 (Core)` - JDK: - `OpenJDK Runtime Environment (build 1.8.0_222-b10)` - `OpenJDK Runtime Environment 18.9 (build 11.0.4+11-LTS)` ### Does this PR introduce any user-facing change? No. ### How was this patch tested? This is a test-only PR. We need to run benchmark. Closes #26003 from dongjoon-hyun/SPARK-29320. Authored-by: Dongjoon Hyun <dhyun@apple.com> Signed-off-by: Dongjoon Hyun <dhyun@apple.com>
2019-10-03 11:58:25 -04:00
year of timestamp: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative
------------------------------------------------------------------------------------------------------------------------
[SPARK-31630][SQL] Fix perf regression by skipping timestamps rebasing after some threshold ### What changes were proposed in this pull request? Skip timestamps rebasing after a global threshold when there is no difference between Julian and Gregorian calendars. This allows to avoid checking hash maps of switch points, and fixes perf regressions in `toJavaTimestamp()` and `fromJavaTimestamp()`. ### Why are the changes needed? The changes fix perf regressions of conversions to/from external type `java.sql.Timestamp`. Before (see the PR's results https://github.com/apache/spark/pull/28440): ``` ================================================================================================ Conversion from/to external types ================================================================================================ OpenJDK 64-Bit Server VM 1.8.0_252-8u252-b09-1~18.04-b09 on Linux 4.15.0-1063-aws Intel(R) Xeon(R) CPU E5-2670 v2 2.50GHz To/from Java's date-time: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative ------------------------------------------------------------------------------------------------------------------------ From java.sql.Timestamp 376 388 10 13.3 75.2 1.1X Collect java.sql.Timestamp 1878 1937 64 2.7 375.6 0.2X ``` After: ``` ================================================================================================ Conversion from/to external types ================================================================================================ OpenJDK 64-Bit Server VM 1.8.0_252-8u252-b09-1~18.04-b09 on Linux 4.15.0-1063-aws Intel(R) Xeon(R) CPU E5-2670 v2 2.50GHz To/from Java's date-time: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative ------------------------------------------------------------------------------------------------------------------------ From java.sql.Timestamp 249 264 24 20.1 49.8 1.7X Collect java.sql.Timestamp 1503 1523 24 3.3 300.5 0.3X ``` Perf improvements in average of: 1. From java.sql.Timestamp is ~ 34% 2. To java.sql.Timestamps is ~16% ### Does this PR introduce _any_ user-facing change? No ### How was this patch tested? By existing test suites `DateTimeUtilsSuite` and `RebaseDateTimeSuite`. Closes #28441 from MaxGekk/opt-rebase-common-threshold. Authored-by: Max Gekk <max.gekk@gmail.com> Signed-off-by: Wenchen Fan <wenchen@databricks.com>
2020-05-05 10:11:53 -04:00
year of timestamp wholestage off 1246 1257 16 8.0 124.6 1.0X
year of timestamp wholestage on 1209 1218 12 8.3 120.9 1.0X
[SPARK-29320][TESTS] Compare `sql/core` module in JDK8/11 (Part 1) ### What changes were proposed in this pull request? This PR regenerates the `sql/core` benchmarks in JDK8/11 to compare the result. In general, we compare the ratio instead of the time. However, in this PR, the average time is compared. This PR should be considered as a rough comparison. **A. EXPECTED CASES(JDK11 is faster in general)** - [x] BloomFilterBenchmark (JDK11 is faster except one case) - [x] BuiltInDataSourceWriteBenchmark (JDK11 is faster at CSV/ORC) - [x] CSVBenchmark (JDK11 is faster except five cases) - [x] ColumnarBatchBenchmark (JDK11 is faster at `boolean`/`string` and some cases in `int`/`array`) - [x] DatasetBenchmark (JDK11 is faster with `string`, but is slower for `long` type) - [x] ExternalAppendOnlyUnsafeRowArrayBenchmark (JDK11 is faster except two cases) - [x] ExtractBenchmark (JDK11 is faster except HOUR/MINUTE/SECOND/MILLISECONDS/MICROSECONDS) - [x] HashedRelationMetricsBenchmark (JDK11 is faster) - [x] JSONBenchmark (JDK11 is much faster except eight cases) - [x] JoinBenchmark (JDK11 is faster except five cases) - [x] OrcNestedSchemaPruningBenchmark (JDK11 is faster in nine cases) - [x] PrimitiveArrayBenchmark (JDK11 is faster) - [x] SortBenchmark (JDK11 is faster except `Arrays.sort` case) - [x] UDFBenchmark (N/A, values are too small) - [x] UnsafeArrayDataBenchmark (JDK11 is faster except one case) - [x] WideTableBenchmark (JDK11 is faster except two cases) **B. CASES WE NEED TO INVESTIGATE MORE LATER** - [x] AggregateBenchmark (JDK11 is slower in general) - [x] CompressionSchemeBenchmark (JDK11 is slower in general except `string`) - [x] DataSourceReadBenchmark (JDK11 is slower in general) - [x] DateTimeBenchmark (JDK11 is slightly slower in general except `parsing`) - [x] MakeDateTimeBenchmark (JDK11 is slower except two cases) - [x] MiscBenchmark (JDK11 is slower except ten cases) - [x] OrcV2NestedSchemaPruningBenchmark (JDK11 is slower) - [x] ParquetNestedSchemaPruningBenchmark (JDK11 is slower except six cases) - [x] RangeBenchmark (JDK11 is slower except one case) `FilterPushdownBenchmark/InExpressionBenchmark/WideSchemaBenchmark` will be compared later because it took long timer. ### Why are the changes needed? According to the result, there are some difference between JDK8/JDK11. This will be a baseline for the future improvement and comparison. Also, as a reproducible environment, the following environment is used. - Instance: `r3.xlarge` - OS: `CentOS Linux release 7.5.1804 (Core)` - JDK: - `OpenJDK Runtime Environment (build 1.8.0_222-b10)` - `OpenJDK Runtime Environment 18.9 (build 11.0.4+11-LTS)` ### Does this PR introduce any user-facing change? No. ### How was this patch tested? This is a test-only PR. We need to run benchmark. Closes #26003 from dongjoon-hyun/SPARK-29320. Authored-by: Dongjoon Hyun <dhyun@apple.com> Signed-off-by: Dongjoon Hyun <dhyun@apple.com>
2019-10-03 11:58:25 -04:00
[SPARK-31527][SQL][TESTS][FOLLOWUP] Fix the number of rows in `DateTimeBenchmark` ### What changes were proposed in this pull request? - Changed to the number of rows in benchmark cases from 3 to the actual number `N`. - Regenerated benchmark results in the environment: | Item | Description | | ---- | ----| | Region | us-west-2 (Oregon) | | Instance | r3.xlarge | | AMI | ubuntu/images/hvm-ssd/ubuntu-bionic-18.04-amd64-server-20190722.1 (ami-06f2f779464715dc5) | | Java | OpenJDK 64-Bit Server VM 1.8.0_242 and OpenJDK 64-Bit Server VM 11.0.6+10 | ### Why are the changes needed? The changes are needed to have: - Correct benchmark results - Base line for other perf improvements that can be checked in the same environment. ### Does this PR introduce _any_ user-facing change? No ### How was this patch tested? By running the benchmark and checking its output. Closes #28440 from MaxGekk/SPARK-31527-DateTimeBenchmark-followup. Authored-by: Max Gekk <max.gekk@gmail.com> Signed-off-by: Takeshi Yamamuro <yamamuro@apache.org>
2020-05-03 20:39:50 -04:00
OpenJDK 64-Bit Server VM 11.0.7+10-post-Ubuntu-2ubuntu218.04 on Linux 4.15.0-1063-aws
Intel(R) Xeon(R) CPU E5-2670 v2 @ 2.50GHz
[SPARK-29320][TESTS] Compare `sql/core` module in JDK8/11 (Part 1) ### What changes were proposed in this pull request? This PR regenerates the `sql/core` benchmarks in JDK8/11 to compare the result. In general, we compare the ratio instead of the time. However, in this PR, the average time is compared. This PR should be considered as a rough comparison. **A. EXPECTED CASES(JDK11 is faster in general)** - [x] BloomFilterBenchmark (JDK11 is faster except one case) - [x] BuiltInDataSourceWriteBenchmark (JDK11 is faster at CSV/ORC) - [x] CSVBenchmark (JDK11 is faster except five cases) - [x] ColumnarBatchBenchmark (JDK11 is faster at `boolean`/`string` and some cases in `int`/`array`) - [x] DatasetBenchmark (JDK11 is faster with `string`, but is slower for `long` type) - [x] ExternalAppendOnlyUnsafeRowArrayBenchmark (JDK11 is faster except two cases) - [x] ExtractBenchmark (JDK11 is faster except HOUR/MINUTE/SECOND/MILLISECONDS/MICROSECONDS) - [x] HashedRelationMetricsBenchmark (JDK11 is faster) - [x] JSONBenchmark (JDK11 is much faster except eight cases) - [x] JoinBenchmark (JDK11 is faster except five cases) - [x] OrcNestedSchemaPruningBenchmark (JDK11 is faster in nine cases) - [x] PrimitiveArrayBenchmark (JDK11 is faster) - [x] SortBenchmark (JDK11 is faster except `Arrays.sort` case) - [x] UDFBenchmark (N/A, values are too small) - [x] UnsafeArrayDataBenchmark (JDK11 is faster except one case) - [x] WideTableBenchmark (JDK11 is faster except two cases) **B. CASES WE NEED TO INVESTIGATE MORE LATER** - [x] AggregateBenchmark (JDK11 is slower in general) - [x] CompressionSchemeBenchmark (JDK11 is slower in general except `string`) - [x] DataSourceReadBenchmark (JDK11 is slower in general) - [x] DateTimeBenchmark (JDK11 is slightly slower in general except `parsing`) - [x] MakeDateTimeBenchmark (JDK11 is slower except two cases) - [x] MiscBenchmark (JDK11 is slower except ten cases) - [x] OrcV2NestedSchemaPruningBenchmark (JDK11 is slower) - [x] ParquetNestedSchemaPruningBenchmark (JDK11 is slower except six cases) - [x] RangeBenchmark (JDK11 is slower except one case) `FilterPushdownBenchmark/InExpressionBenchmark/WideSchemaBenchmark` will be compared later because it took long timer. ### Why are the changes needed? According to the result, there are some difference between JDK8/JDK11. This will be a baseline for the future improvement and comparison. Also, as a reproducible environment, the following environment is used. - Instance: `r3.xlarge` - OS: `CentOS Linux release 7.5.1804 (Core)` - JDK: - `OpenJDK Runtime Environment (build 1.8.0_222-b10)` - `OpenJDK Runtime Environment 18.9 (build 11.0.4+11-LTS)` ### Does this PR introduce any user-facing change? No. ### How was this patch tested? This is a test-only PR. We need to run benchmark. Closes #26003 from dongjoon-hyun/SPARK-29320. Authored-by: Dongjoon Hyun <dhyun@apple.com> Signed-off-by: Dongjoon Hyun <dhyun@apple.com>
2019-10-03 11:58:25 -04:00
quarter of timestamp: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative
------------------------------------------------------------------------------------------------------------------------
[SPARK-31630][SQL] Fix perf regression by skipping timestamps rebasing after some threshold ### What changes were proposed in this pull request? Skip timestamps rebasing after a global threshold when there is no difference between Julian and Gregorian calendars. This allows to avoid checking hash maps of switch points, and fixes perf regressions in `toJavaTimestamp()` and `fromJavaTimestamp()`. ### Why are the changes needed? The changes fix perf regressions of conversions to/from external type `java.sql.Timestamp`. Before (see the PR's results https://github.com/apache/spark/pull/28440): ``` ================================================================================================ Conversion from/to external types ================================================================================================ OpenJDK 64-Bit Server VM 1.8.0_252-8u252-b09-1~18.04-b09 on Linux 4.15.0-1063-aws Intel(R) Xeon(R) CPU E5-2670 v2 2.50GHz To/from Java's date-time: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative ------------------------------------------------------------------------------------------------------------------------ From java.sql.Timestamp 376 388 10 13.3 75.2 1.1X Collect java.sql.Timestamp 1878 1937 64 2.7 375.6 0.2X ``` After: ``` ================================================================================================ Conversion from/to external types ================================================================================================ OpenJDK 64-Bit Server VM 1.8.0_252-8u252-b09-1~18.04-b09 on Linux 4.15.0-1063-aws Intel(R) Xeon(R) CPU E5-2670 v2 2.50GHz To/from Java's date-time: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative ------------------------------------------------------------------------------------------------------------------------ From java.sql.Timestamp 249 264 24 20.1 49.8 1.7X Collect java.sql.Timestamp 1503 1523 24 3.3 300.5 0.3X ``` Perf improvements in average of: 1. From java.sql.Timestamp is ~ 34% 2. To java.sql.Timestamps is ~16% ### Does this PR introduce _any_ user-facing change? No ### How was this patch tested? By existing test suites `DateTimeUtilsSuite` and `RebaseDateTimeSuite`. Closes #28441 from MaxGekk/opt-rebase-common-threshold. Authored-by: Max Gekk <max.gekk@gmail.com> Signed-off-by: Wenchen Fan <wenchen@databricks.com>
2020-05-05 10:11:53 -04:00
quarter of timestamp wholestage off 1608 1616 11 6.2 160.8 1.0X
quarter of timestamp wholestage on 1540 1552 10 6.5 154.0 1.0X
[SPARK-29320][TESTS] Compare `sql/core` module in JDK8/11 (Part 1) ### What changes were proposed in this pull request? This PR regenerates the `sql/core` benchmarks in JDK8/11 to compare the result. In general, we compare the ratio instead of the time. However, in this PR, the average time is compared. This PR should be considered as a rough comparison. **A. EXPECTED CASES(JDK11 is faster in general)** - [x] BloomFilterBenchmark (JDK11 is faster except one case) - [x] BuiltInDataSourceWriteBenchmark (JDK11 is faster at CSV/ORC) - [x] CSVBenchmark (JDK11 is faster except five cases) - [x] ColumnarBatchBenchmark (JDK11 is faster at `boolean`/`string` and some cases in `int`/`array`) - [x] DatasetBenchmark (JDK11 is faster with `string`, but is slower for `long` type) - [x] ExternalAppendOnlyUnsafeRowArrayBenchmark (JDK11 is faster except two cases) - [x] ExtractBenchmark (JDK11 is faster except HOUR/MINUTE/SECOND/MILLISECONDS/MICROSECONDS) - [x] HashedRelationMetricsBenchmark (JDK11 is faster) - [x] JSONBenchmark (JDK11 is much faster except eight cases) - [x] JoinBenchmark (JDK11 is faster except five cases) - [x] OrcNestedSchemaPruningBenchmark (JDK11 is faster in nine cases) - [x] PrimitiveArrayBenchmark (JDK11 is faster) - [x] SortBenchmark (JDK11 is faster except `Arrays.sort` case) - [x] UDFBenchmark (N/A, values are too small) - [x] UnsafeArrayDataBenchmark (JDK11 is faster except one case) - [x] WideTableBenchmark (JDK11 is faster except two cases) **B. CASES WE NEED TO INVESTIGATE MORE LATER** - [x] AggregateBenchmark (JDK11 is slower in general) - [x] CompressionSchemeBenchmark (JDK11 is slower in general except `string`) - [x] DataSourceReadBenchmark (JDK11 is slower in general) - [x] DateTimeBenchmark (JDK11 is slightly slower in general except `parsing`) - [x] MakeDateTimeBenchmark (JDK11 is slower except two cases) - [x] MiscBenchmark (JDK11 is slower except ten cases) - [x] OrcV2NestedSchemaPruningBenchmark (JDK11 is slower) - [x] ParquetNestedSchemaPruningBenchmark (JDK11 is slower except six cases) - [x] RangeBenchmark (JDK11 is slower except one case) `FilterPushdownBenchmark/InExpressionBenchmark/WideSchemaBenchmark` will be compared later because it took long timer. ### Why are the changes needed? According to the result, there are some difference between JDK8/JDK11. This will be a baseline for the future improvement and comparison. Also, as a reproducible environment, the following environment is used. - Instance: `r3.xlarge` - OS: `CentOS Linux release 7.5.1804 (Core)` - JDK: - `OpenJDK Runtime Environment (build 1.8.0_222-b10)` - `OpenJDK Runtime Environment 18.9 (build 11.0.4+11-LTS)` ### Does this PR introduce any user-facing change? No. ### How was this patch tested? This is a test-only PR. We need to run benchmark. Closes #26003 from dongjoon-hyun/SPARK-29320. Authored-by: Dongjoon Hyun <dhyun@apple.com> Signed-off-by: Dongjoon Hyun <dhyun@apple.com>
2019-10-03 11:58:25 -04:00
[SPARK-31527][SQL][TESTS][FOLLOWUP] Fix the number of rows in `DateTimeBenchmark` ### What changes were proposed in this pull request? - Changed to the number of rows in benchmark cases from 3 to the actual number `N`. - Regenerated benchmark results in the environment: | Item | Description | | ---- | ----| | Region | us-west-2 (Oregon) | | Instance | r3.xlarge | | AMI | ubuntu/images/hvm-ssd/ubuntu-bionic-18.04-amd64-server-20190722.1 (ami-06f2f779464715dc5) | | Java | OpenJDK 64-Bit Server VM 1.8.0_242 and OpenJDK 64-Bit Server VM 11.0.6+10 | ### Why are the changes needed? The changes are needed to have: - Correct benchmark results - Base line for other perf improvements that can be checked in the same environment. ### Does this PR introduce _any_ user-facing change? No ### How was this patch tested? By running the benchmark and checking its output. Closes #28440 from MaxGekk/SPARK-31527-DateTimeBenchmark-followup. Authored-by: Max Gekk <max.gekk@gmail.com> Signed-off-by: Takeshi Yamamuro <yamamuro@apache.org>
2020-05-03 20:39:50 -04:00
OpenJDK 64-Bit Server VM 11.0.7+10-post-Ubuntu-2ubuntu218.04 on Linux 4.15.0-1063-aws
Intel(R) Xeon(R) CPU E5-2670 v2 @ 2.50GHz
[SPARK-29320][TESTS] Compare `sql/core` module in JDK8/11 (Part 1) ### What changes were proposed in this pull request? This PR regenerates the `sql/core` benchmarks in JDK8/11 to compare the result. In general, we compare the ratio instead of the time. However, in this PR, the average time is compared. This PR should be considered as a rough comparison. **A. EXPECTED CASES(JDK11 is faster in general)** - [x] BloomFilterBenchmark (JDK11 is faster except one case) - [x] BuiltInDataSourceWriteBenchmark (JDK11 is faster at CSV/ORC) - [x] CSVBenchmark (JDK11 is faster except five cases) - [x] ColumnarBatchBenchmark (JDK11 is faster at `boolean`/`string` and some cases in `int`/`array`) - [x] DatasetBenchmark (JDK11 is faster with `string`, but is slower for `long` type) - [x] ExternalAppendOnlyUnsafeRowArrayBenchmark (JDK11 is faster except two cases) - [x] ExtractBenchmark (JDK11 is faster except HOUR/MINUTE/SECOND/MILLISECONDS/MICROSECONDS) - [x] HashedRelationMetricsBenchmark (JDK11 is faster) - [x] JSONBenchmark (JDK11 is much faster except eight cases) - [x] JoinBenchmark (JDK11 is faster except five cases) - [x] OrcNestedSchemaPruningBenchmark (JDK11 is faster in nine cases) - [x] PrimitiveArrayBenchmark (JDK11 is faster) - [x] SortBenchmark (JDK11 is faster except `Arrays.sort` case) - [x] UDFBenchmark (N/A, values are too small) - [x] UnsafeArrayDataBenchmark (JDK11 is faster except one case) - [x] WideTableBenchmark (JDK11 is faster except two cases) **B. CASES WE NEED TO INVESTIGATE MORE LATER** - [x] AggregateBenchmark (JDK11 is slower in general) - [x] CompressionSchemeBenchmark (JDK11 is slower in general except `string`) - [x] DataSourceReadBenchmark (JDK11 is slower in general) - [x] DateTimeBenchmark (JDK11 is slightly slower in general except `parsing`) - [x] MakeDateTimeBenchmark (JDK11 is slower except two cases) - [x] MiscBenchmark (JDK11 is slower except ten cases) - [x] OrcV2NestedSchemaPruningBenchmark (JDK11 is slower) - [x] ParquetNestedSchemaPruningBenchmark (JDK11 is slower except six cases) - [x] RangeBenchmark (JDK11 is slower except one case) `FilterPushdownBenchmark/InExpressionBenchmark/WideSchemaBenchmark` will be compared later because it took long timer. ### Why are the changes needed? According to the result, there are some difference between JDK8/JDK11. This will be a baseline for the future improvement and comparison. Also, as a reproducible environment, the following environment is used. - Instance: `r3.xlarge` - OS: `CentOS Linux release 7.5.1804 (Core)` - JDK: - `OpenJDK Runtime Environment (build 1.8.0_222-b10)` - `OpenJDK Runtime Environment 18.9 (build 11.0.4+11-LTS)` ### Does this PR introduce any user-facing change? No. ### How was this patch tested? This is a test-only PR. We need to run benchmark. Closes #26003 from dongjoon-hyun/SPARK-29320. Authored-by: Dongjoon Hyun <dhyun@apple.com> Signed-off-by: Dongjoon Hyun <dhyun@apple.com>
2019-10-03 11:58:25 -04:00
month of timestamp: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative
------------------------------------------------------------------------------------------------------------------------
[SPARK-31630][SQL] Fix perf regression by skipping timestamps rebasing after some threshold ### What changes were proposed in this pull request? Skip timestamps rebasing after a global threshold when there is no difference between Julian and Gregorian calendars. This allows to avoid checking hash maps of switch points, and fixes perf regressions in `toJavaTimestamp()` and `fromJavaTimestamp()`. ### Why are the changes needed? The changes fix perf regressions of conversions to/from external type `java.sql.Timestamp`. Before (see the PR's results https://github.com/apache/spark/pull/28440): ``` ================================================================================================ Conversion from/to external types ================================================================================================ OpenJDK 64-Bit Server VM 1.8.0_252-8u252-b09-1~18.04-b09 on Linux 4.15.0-1063-aws Intel(R) Xeon(R) CPU E5-2670 v2 2.50GHz To/from Java's date-time: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative ------------------------------------------------------------------------------------------------------------------------ From java.sql.Timestamp 376 388 10 13.3 75.2 1.1X Collect java.sql.Timestamp 1878 1937 64 2.7 375.6 0.2X ``` After: ``` ================================================================================================ Conversion from/to external types ================================================================================================ OpenJDK 64-Bit Server VM 1.8.0_252-8u252-b09-1~18.04-b09 on Linux 4.15.0-1063-aws Intel(R) Xeon(R) CPU E5-2670 v2 2.50GHz To/from Java's date-time: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative ------------------------------------------------------------------------------------------------------------------------ From java.sql.Timestamp 249 264 24 20.1 49.8 1.7X Collect java.sql.Timestamp 1503 1523 24 3.3 300.5 0.3X ``` Perf improvements in average of: 1. From java.sql.Timestamp is ~ 34% 2. To java.sql.Timestamps is ~16% ### Does this PR introduce _any_ user-facing change? No ### How was this patch tested? By existing test suites `DateTimeUtilsSuite` and `RebaseDateTimeSuite`. Closes #28441 from MaxGekk/opt-rebase-common-threshold. Authored-by: Max Gekk <max.gekk@gmail.com> Signed-off-by: Wenchen Fan <wenchen@databricks.com>
2020-05-05 10:11:53 -04:00
month of timestamp wholestage off 1242 1246 6 8.1 124.2 1.0X
month of timestamp wholestage on 1202 1212 11 8.3 120.2 1.0X
[SPARK-29320][TESTS] Compare `sql/core` module in JDK8/11 (Part 1) ### What changes were proposed in this pull request? This PR regenerates the `sql/core` benchmarks in JDK8/11 to compare the result. In general, we compare the ratio instead of the time. However, in this PR, the average time is compared. This PR should be considered as a rough comparison. **A. EXPECTED CASES(JDK11 is faster in general)** - [x] BloomFilterBenchmark (JDK11 is faster except one case) - [x] BuiltInDataSourceWriteBenchmark (JDK11 is faster at CSV/ORC) - [x] CSVBenchmark (JDK11 is faster except five cases) - [x] ColumnarBatchBenchmark (JDK11 is faster at `boolean`/`string` and some cases in `int`/`array`) - [x] DatasetBenchmark (JDK11 is faster with `string`, but is slower for `long` type) - [x] ExternalAppendOnlyUnsafeRowArrayBenchmark (JDK11 is faster except two cases) - [x] ExtractBenchmark (JDK11 is faster except HOUR/MINUTE/SECOND/MILLISECONDS/MICROSECONDS) - [x] HashedRelationMetricsBenchmark (JDK11 is faster) - [x] JSONBenchmark (JDK11 is much faster except eight cases) - [x] JoinBenchmark (JDK11 is faster except five cases) - [x] OrcNestedSchemaPruningBenchmark (JDK11 is faster in nine cases) - [x] PrimitiveArrayBenchmark (JDK11 is faster) - [x] SortBenchmark (JDK11 is faster except `Arrays.sort` case) - [x] UDFBenchmark (N/A, values are too small) - [x] UnsafeArrayDataBenchmark (JDK11 is faster except one case) - [x] WideTableBenchmark (JDK11 is faster except two cases) **B. CASES WE NEED TO INVESTIGATE MORE LATER** - [x] AggregateBenchmark (JDK11 is slower in general) - [x] CompressionSchemeBenchmark (JDK11 is slower in general except `string`) - [x] DataSourceReadBenchmark (JDK11 is slower in general) - [x] DateTimeBenchmark (JDK11 is slightly slower in general except `parsing`) - [x] MakeDateTimeBenchmark (JDK11 is slower except two cases) - [x] MiscBenchmark (JDK11 is slower except ten cases) - [x] OrcV2NestedSchemaPruningBenchmark (JDK11 is slower) - [x] ParquetNestedSchemaPruningBenchmark (JDK11 is slower except six cases) - [x] RangeBenchmark (JDK11 is slower except one case) `FilterPushdownBenchmark/InExpressionBenchmark/WideSchemaBenchmark` will be compared later because it took long timer. ### Why are the changes needed? According to the result, there are some difference between JDK8/JDK11. This will be a baseline for the future improvement and comparison. Also, as a reproducible environment, the following environment is used. - Instance: `r3.xlarge` - OS: `CentOS Linux release 7.5.1804 (Core)` - JDK: - `OpenJDK Runtime Environment (build 1.8.0_222-b10)` - `OpenJDK Runtime Environment 18.9 (build 11.0.4+11-LTS)` ### Does this PR introduce any user-facing change? No. ### How was this patch tested? This is a test-only PR. We need to run benchmark. Closes #26003 from dongjoon-hyun/SPARK-29320. Authored-by: Dongjoon Hyun <dhyun@apple.com> Signed-off-by: Dongjoon Hyun <dhyun@apple.com>
2019-10-03 11:58:25 -04:00
[SPARK-31527][SQL][TESTS][FOLLOWUP] Fix the number of rows in `DateTimeBenchmark` ### What changes were proposed in this pull request? - Changed to the number of rows in benchmark cases from 3 to the actual number `N`. - Regenerated benchmark results in the environment: | Item | Description | | ---- | ----| | Region | us-west-2 (Oregon) | | Instance | r3.xlarge | | AMI | ubuntu/images/hvm-ssd/ubuntu-bionic-18.04-amd64-server-20190722.1 (ami-06f2f779464715dc5) | | Java | OpenJDK 64-Bit Server VM 1.8.0_242 and OpenJDK 64-Bit Server VM 11.0.6+10 | ### Why are the changes needed? The changes are needed to have: - Correct benchmark results - Base line for other perf improvements that can be checked in the same environment. ### Does this PR introduce _any_ user-facing change? No ### How was this patch tested? By running the benchmark and checking its output. Closes #28440 from MaxGekk/SPARK-31527-DateTimeBenchmark-followup. Authored-by: Max Gekk <max.gekk@gmail.com> Signed-off-by: Takeshi Yamamuro <yamamuro@apache.org>
2020-05-03 20:39:50 -04:00
OpenJDK 64-Bit Server VM 11.0.7+10-post-Ubuntu-2ubuntu218.04 on Linux 4.15.0-1063-aws
Intel(R) Xeon(R) CPU E5-2670 v2 @ 2.50GHz
[SPARK-29320][TESTS] Compare `sql/core` module in JDK8/11 (Part 1) ### What changes were proposed in this pull request? This PR regenerates the `sql/core` benchmarks in JDK8/11 to compare the result. In general, we compare the ratio instead of the time. However, in this PR, the average time is compared. This PR should be considered as a rough comparison. **A. EXPECTED CASES(JDK11 is faster in general)** - [x] BloomFilterBenchmark (JDK11 is faster except one case) - [x] BuiltInDataSourceWriteBenchmark (JDK11 is faster at CSV/ORC) - [x] CSVBenchmark (JDK11 is faster except five cases) - [x] ColumnarBatchBenchmark (JDK11 is faster at `boolean`/`string` and some cases in `int`/`array`) - [x] DatasetBenchmark (JDK11 is faster with `string`, but is slower for `long` type) - [x] ExternalAppendOnlyUnsafeRowArrayBenchmark (JDK11 is faster except two cases) - [x] ExtractBenchmark (JDK11 is faster except HOUR/MINUTE/SECOND/MILLISECONDS/MICROSECONDS) - [x] HashedRelationMetricsBenchmark (JDK11 is faster) - [x] JSONBenchmark (JDK11 is much faster except eight cases) - [x] JoinBenchmark (JDK11 is faster except five cases) - [x] OrcNestedSchemaPruningBenchmark (JDK11 is faster in nine cases) - [x] PrimitiveArrayBenchmark (JDK11 is faster) - [x] SortBenchmark (JDK11 is faster except `Arrays.sort` case) - [x] UDFBenchmark (N/A, values are too small) - [x] UnsafeArrayDataBenchmark (JDK11 is faster except one case) - [x] WideTableBenchmark (JDK11 is faster except two cases) **B. CASES WE NEED TO INVESTIGATE MORE LATER** - [x] AggregateBenchmark (JDK11 is slower in general) - [x] CompressionSchemeBenchmark (JDK11 is slower in general except `string`) - [x] DataSourceReadBenchmark (JDK11 is slower in general) - [x] DateTimeBenchmark (JDK11 is slightly slower in general except `parsing`) - [x] MakeDateTimeBenchmark (JDK11 is slower except two cases) - [x] MiscBenchmark (JDK11 is slower except ten cases) - [x] OrcV2NestedSchemaPruningBenchmark (JDK11 is slower) - [x] ParquetNestedSchemaPruningBenchmark (JDK11 is slower except six cases) - [x] RangeBenchmark (JDK11 is slower except one case) `FilterPushdownBenchmark/InExpressionBenchmark/WideSchemaBenchmark` will be compared later because it took long timer. ### Why are the changes needed? According to the result, there are some difference between JDK8/JDK11. This will be a baseline for the future improvement and comparison. Also, as a reproducible environment, the following environment is used. - Instance: `r3.xlarge` - OS: `CentOS Linux release 7.5.1804 (Core)` - JDK: - `OpenJDK Runtime Environment (build 1.8.0_222-b10)` - `OpenJDK Runtime Environment 18.9 (build 11.0.4+11-LTS)` ### Does this PR introduce any user-facing change? No. ### How was this patch tested? This is a test-only PR. We need to run benchmark. Closes #26003 from dongjoon-hyun/SPARK-29320. Authored-by: Dongjoon Hyun <dhyun@apple.com> Signed-off-by: Dongjoon Hyun <dhyun@apple.com>
2019-10-03 11:58:25 -04:00
weekofyear of timestamp: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative
------------------------------------------------------------------------------------------------------------------------
[SPARK-31630][SQL] Fix perf regression by skipping timestamps rebasing after some threshold ### What changes were proposed in this pull request? Skip timestamps rebasing after a global threshold when there is no difference between Julian and Gregorian calendars. This allows to avoid checking hash maps of switch points, and fixes perf regressions in `toJavaTimestamp()` and `fromJavaTimestamp()`. ### Why are the changes needed? The changes fix perf regressions of conversions to/from external type `java.sql.Timestamp`. Before (see the PR's results https://github.com/apache/spark/pull/28440): ``` ================================================================================================ Conversion from/to external types ================================================================================================ OpenJDK 64-Bit Server VM 1.8.0_252-8u252-b09-1~18.04-b09 on Linux 4.15.0-1063-aws Intel(R) Xeon(R) CPU E5-2670 v2 2.50GHz To/from Java's date-time: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative ------------------------------------------------------------------------------------------------------------------------ From java.sql.Timestamp 376 388 10 13.3 75.2 1.1X Collect java.sql.Timestamp 1878 1937 64 2.7 375.6 0.2X ``` After: ``` ================================================================================================ Conversion from/to external types ================================================================================================ OpenJDK 64-Bit Server VM 1.8.0_252-8u252-b09-1~18.04-b09 on Linux 4.15.0-1063-aws Intel(R) Xeon(R) CPU E5-2670 v2 2.50GHz To/from Java's date-time: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative ------------------------------------------------------------------------------------------------------------------------ From java.sql.Timestamp 249 264 24 20.1 49.8 1.7X Collect java.sql.Timestamp 1503 1523 24 3.3 300.5 0.3X ``` Perf improvements in average of: 1. From java.sql.Timestamp is ~ 34% 2. To java.sql.Timestamps is ~16% ### Does this PR introduce _any_ user-facing change? No ### How was this patch tested? By existing test suites `DateTimeUtilsSuite` and `RebaseDateTimeSuite`. Closes #28441 from MaxGekk/opt-rebase-common-threshold. Authored-by: Max Gekk <max.gekk@gmail.com> Signed-off-by: Wenchen Fan <wenchen@databricks.com>
2020-05-05 10:11:53 -04:00
weekofyear of timestamp wholestage off 1879 1885 8 5.3 187.9 1.0X
weekofyear of timestamp wholestage on 1832 1845 10 5.5 183.2 1.0X
[SPARK-29320][TESTS] Compare `sql/core` module in JDK8/11 (Part 1) ### What changes were proposed in this pull request? This PR regenerates the `sql/core` benchmarks in JDK8/11 to compare the result. In general, we compare the ratio instead of the time. However, in this PR, the average time is compared. This PR should be considered as a rough comparison. **A. EXPECTED CASES(JDK11 is faster in general)** - [x] BloomFilterBenchmark (JDK11 is faster except one case) - [x] BuiltInDataSourceWriteBenchmark (JDK11 is faster at CSV/ORC) - [x] CSVBenchmark (JDK11 is faster except five cases) - [x] ColumnarBatchBenchmark (JDK11 is faster at `boolean`/`string` and some cases in `int`/`array`) - [x] DatasetBenchmark (JDK11 is faster with `string`, but is slower for `long` type) - [x] ExternalAppendOnlyUnsafeRowArrayBenchmark (JDK11 is faster except two cases) - [x] ExtractBenchmark (JDK11 is faster except HOUR/MINUTE/SECOND/MILLISECONDS/MICROSECONDS) - [x] HashedRelationMetricsBenchmark (JDK11 is faster) - [x] JSONBenchmark (JDK11 is much faster except eight cases) - [x] JoinBenchmark (JDK11 is faster except five cases) - [x] OrcNestedSchemaPruningBenchmark (JDK11 is faster in nine cases) - [x] PrimitiveArrayBenchmark (JDK11 is faster) - [x] SortBenchmark (JDK11 is faster except `Arrays.sort` case) - [x] UDFBenchmark (N/A, values are too small) - [x] UnsafeArrayDataBenchmark (JDK11 is faster except one case) - [x] WideTableBenchmark (JDK11 is faster except two cases) **B. CASES WE NEED TO INVESTIGATE MORE LATER** - [x] AggregateBenchmark (JDK11 is slower in general) - [x] CompressionSchemeBenchmark (JDK11 is slower in general except `string`) - [x] DataSourceReadBenchmark (JDK11 is slower in general) - [x] DateTimeBenchmark (JDK11 is slightly slower in general except `parsing`) - [x] MakeDateTimeBenchmark (JDK11 is slower except two cases) - [x] MiscBenchmark (JDK11 is slower except ten cases) - [x] OrcV2NestedSchemaPruningBenchmark (JDK11 is slower) - [x] ParquetNestedSchemaPruningBenchmark (JDK11 is slower except six cases) - [x] RangeBenchmark (JDK11 is slower except one case) `FilterPushdownBenchmark/InExpressionBenchmark/WideSchemaBenchmark` will be compared later because it took long timer. ### Why are the changes needed? According to the result, there are some difference between JDK8/JDK11. This will be a baseline for the future improvement and comparison. Also, as a reproducible environment, the following environment is used. - Instance: `r3.xlarge` - OS: `CentOS Linux release 7.5.1804 (Core)` - JDK: - `OpenJDK Runtime Environment (build 1.8.0_222-b10)` - `OpenJDK Runtime Environment 18.9 (build 11.0.4+11-LTS)` ### Does this PR introduce any user-facing change? No. ### How was this patch tested? This is a test-only PR. We need to run benchmark. Closes #26003 from dongjoon-hyun/SPARK-29320. Authored-by: Dongjoon Hyun <dhyun@apple.com> Signed-off-by: Dongjoon Hyun <dhyun@apple.com>
2019-10-03 11:58:25 -04:00
[SPARK-31527][SQL][TESTS][FOLLOWUP] Fix the number of rows in `DateTimeBenchmark` ### What changes were proposed in this pull request? - Changed to the number of rows in benchmark cases from 3 to the actual number `N`. - Regenerated benchmark results in the environment: | Item | Description | | ---- | ----| | Region | us-west-2 (Oregon) | | Instance | r3.xlarge | | AMI | ubuntu/images/hvm-ssd/ubuntu-bionic-18.04-amd64-server-20190722.1 (ami-06f2f779464715dc5) | | Java | OpenJDK 64-Bit Server VM 1.8.0_242 and OpenJDK 64-Bit Server VM 11.0.6+10 | ### Why are the changes needed? The changes are needed to have: - Correct benchmark results - Base line for other perf improvements that can be checked in the same environment. ### Does this PR introduce _any_ user-facing change? No ### How was this patch tested? By running the benchmark and checking its output. Closes #28440 from MaxGekk/SPARK-31527-DateTimeBenchmark-followup. Authored-by: Max Gekk <max.gekk@gmail.com> Signed-off-by: Takeshi Yamamuro <yamamuro@apache.org>
2020-05-03 20:39:50 -04:00
OpenJDK 64-Bit Server VM 11.0.7+10-post-Ubuntu-2ubuntu218.04 on Linux 4.15.0-1063-aws
Intel(R) Xeon(R) CPU E5-2670 v2 @ 2.50GHz
[SPARK-29320][TESTS] Compare `sql/core` module in JDK8/11 (Part 1) ### What changes were proposed in this pull request? This PR regenerates the `sql/core` benchmarks in JDK8/11 to compare the result. In general, we compare the ratio instead of the time. However, in this PR, the average time is compared. This PR should be considered as a rough comparison. **A. EXPECTED CASES(JDK11 is faster in general)** - [x] BloomFilterBenchmark (JDK11 is faster except one case) - [x] BuiltInDataSourceWriteBenchmark (JDK11 is faster at CSV/ORC) - [x] CSVBenchmark (JDK11 is faster except five cases) - [x] ColumnarBatchBenchmark (JDK11 is faster at `boolean`/`string` and some cases in `int`/`array`) - [x] DatasetBenchmark (JDK11 is faster with `string`, but is slower for `long` type) - [x] ExternalAppendOnlyUnsafeRowArrayBenchmark (JDK11 is faster except two cases) - [x] ExtractBenchmark (JDK11 is faster except HOUR/MINUTE/SECOND/MILLISECONDS/MICROSECONDS) - [x] HashedRelationMetricsBenchmark (JDK11 is faster) - [x] JSONBenchmark (JDK11 is much faster except eight cases) - [x] JoinBenchmark (JDK11 is faster except five cases) - [x] OrcNestedSchemaPruningBenchmark (JDK11 is faster in nine cases) - [x] PrimitiveArrayBenchmark (JDK11 is faster) - [x] SortBenchmark (JDK11 is faster except `Arrays.sort` case) - [x] UDFBenchmark (N/A, values are too small) - [x] UnsafeArrayDataBenchmark (JDK11 is faster except one case) - [x] WideTableBenchmark (JDK11 is faster except two cases) **B. CASES WE NEED TO INVESTIGATE MORE LATER** - [x] AggregateBenchmark (JDK11 is slower in general) - [x] CompressionSchemeBenchmark (JDK11 is slower in general except `string`) - [x] DataSourceReadBenchmark (JDK11 is slower in general) - [x] DateTimeBenchmark (JDK11 is slightly slower in general except `parsing`) - [x] MakeDateTimeBenchmark (JDK11 is slower except two cases) - [x] MiscBenchmark (JDK11 is slower except ten cases) - [x] OrcV2NestedSchemaPruningBenchmark (JDK11 is slower) - [x] ParquetNestedSchemaPruningBenchmark (JDK11 is slower except six cases) - [x] RangeBenchmark (JDK11 is slower except one case) `FilterPushdownBenchmark/InExpressionBenchmark/WideSchemaBenchmark` will be compared later because it took long timer. ### Why are the changes needed? According to the result, there are some difference between JDK8/JDK11. This will be a baseline for the future improvement and comparison. Also, as a reproducible environment, the following environment is used. - Instance: `r3.xlarge` - OS: `CentOS Linux release 7.5.1804 (Core)` - JDK: - `OpenJDK Runtime Environment (build 1.8.0_222-b10)` - `OpenJDK Runtime Environment 18.9 (build 11.0.4+11-LTS)` ### Does this PR introduce any user-facing change? No. ### How was this patch tested? This is a test-only PR. We need to run benchmark. Closes #26003 from dongjoon-hyun/SPARK-29320. Authored-by: Dongjoon Hyun <dhyun@apple.com> Signed-off-by: Dongjoon Hyun <dhyun@apple.com>
2019-10-03 11:58:25 -04:00
day of timestamp: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative
------------------------------------------------------------------------------------------------------------------------
[SPARK-31630][SQL] Fix perf regression by skipping timestamps rebasing after some threshold ### What changes were proposed in this pull request? Skip timestamps rebasing after a global threshold when there is no difference between Julian and Gregorian calendars. This allows to avoid checking hash maps of switch points, and fixes perf regressions in `toJavaTimestamp()` and `fromJavaTimestamp()`. ### Why are the changes needed? The changes fix perf regressions of conversions to/from external type `java.sql.Timestamp`. Before (see the PR's results https://github.com/apache/spark/pull/28440): ``` ================================================================================================ Conversion from/to external types ================================================================================================ OpenJDK 64-Bit Server VM 1.8.0_252-8u252-b09-1~18.04-b09 on Linux 4.15.0-1063-aws Intel(R) Xeon(R) CPU E5-2670 v2 2.50GHz To/from Java's date-time: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative ------------------------------------------------------------------------------------------------------------------------ From java.sql.Timestamp 376 388 10 13.3 75.2 1.1X Collect java.sql.Timestamp 1878 1937 64 2.7 375.6 0.2X ``` After: ``` ================================================================================================ Conversion from/to external types ================================================================================================ OpenJDK 64-Bit Server VM 1.8.0_252-8u252-b09-1~18.04-b09 on Linux 4.15.0-1063-aws Intel(R) Xeon(R) CPU E5-2670 v2 2.50GHz To/from Java's date-time: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative ------------------------------------------------------------------------------------------------------------------------ From java.sql.Timestamp 249 264 24 20.1 49.8 1.7X Collect java.sql.Timestamp 1503 1523 24 3.3 300.5 0.3X ``` Perf improvements in average of: 1. From java.sql.Timestamp is ~ 34% 2. To java.sql.Timestamps is ~16% ### Does this PR introduce _any_ user-facing change? No ### How was this patch tested? By existing test suites `DateTimeUtilsSuite` and `RebaseDateTimeSuite`. Closes #28441 from MaxGekk/opt-rebase-common-threshold. Authored-by: Max Gekk <max.gekk@gmail.com> Signed-off-by: Wenchen Fan <wenchen@databricks.com>
2020-05-05 10:11:53 -04:00
day of timestamp wholestage off 1236 1239 4 8.1 123.6 1.0X
day of timestamp wholestage on 1206 1219 17 8.3 120.6 1.0X
[SPARK-29320][TESTS] Compare `sql/core` module in JDK8/11 (Part 1) ### What changes were proposed in this pull request? This PR regenerates the `sql/core` benchmarks in JDK8/11 to compare the result. In general, we compare the ratio instead of the time. However, in this PR, the average time is compared. This PR should be considered as a rough comparison. **A. EXPECTED CASES(JDK11 is faster in general)** - [x] BloomFilterBenchmark (JDK11 is faster except one case) - [x] BuiltInDataSourceWriteBenchmark (JDK11 is faster at CSV/ORC) - [x] CSVBenchmark (JDK11 is faster except five cases) - [x] ColumnarBatchBenchmark (JDK11 is faster at `boolean`/`string` and some cases in `int`/`array`) - [x] DatasetBenchmark (JDK11 is faster with `string`, but is slower for `long` type) - [x] ExternalAppendOnlyUnsafeRowArrayBenchmark (JDK11 is faster except two cases) - [x] ExtractBenchmark (JDK11 is faster except HOUR/MINUTE/SECOND/MILLISECONDS/MICROSECONDS) - [x] HashedRelationMetricsBenchmark (JDK11 is faster) - [x] JSONBenchmark (JDK11 is much faster except eight cases) - [x] JoinBenchmark (JDK11 is faster except five cases) - [x] OrcNestedSchemaPruningBenchmark (JDK11 is faster in nine cases) - [x] PrimitiveArrayBenchmark (JDK11 is faster) - [x] SortBenchmark (JDK11 is faster except `Arrays.sort` case) - [x] UDFBenchmark (N/A, values are too small) - [x] UnsafeArrayDataBenchmark (JDK11 is faster except one case) - [x] WideTableBenchmark (JDK11 is faster except two cases) **B. CASES WE NEED TO INVESTIGATE MORE LATER** - [x] AggregateBenchmark (JDK11 is slower in general) - [x] CompressionSchemeBenchmark (JDK11 is slower in general except `string`) - [x] DataSourceReadBenchmark (JDK11 is slower in general) - [x] DateTimeBenchmark (JDK11 is slightly slower in general except `parsing`) - [x] MakeDateTimeBenchmark (JDK11 is slower except two cases) - [x] MiscBenchmark (JDK11 is slower except ten cases) - [x] OrcV2NestedSchemaPruningBenchmark (JDK11 is slower) - [x] ParquetNestedSchemaPruningBenchmark (JDK11 is slower except six cases) - [x] RangeBenchmark (JDK11 is slower except one case) `FilterPushdownBenchmark/InExpressionBenchmark/WideSchemaBenchmark` will be compared later because it took long timer. ### Why are the changes needed? According to the result, there are some difference between JDK8/JDK11. This will be a baseline for the future improvement and comparison. Also, as a reproducible environment, the following environment is used. - Instance: `r3.xlarge` - OS: `CentOS Linux release 7.5.1804 (Core)` - JDK: - `OpenJDK Runtime Environment (build 1.8.0_222-b10)` - `OpenJDK Runtime Environment 18.9 (build 11.0.4+11-LTS)` ### Does this PR introduce any user-facing change? No. ### How was this patch tested? This is a test-only PR. We need to run benchmark. Closes #26003 from dongjoon-hyun/SPARK-29320. Authored-by: Dongjoon Hyun <dhyun@apple.com> Signed-off-by: Dongjoon Hyun <dhyun@apple.com>
2019-10-03 11:58:25 -04:00
[SPARK-31527][SQL][TESTS][FOLLOWUP] Fix the number of rows in `DateTimeBenchmark` ### What changes were proposed in this pull request? - Changed to the number of rows in benchmark cases from 3 to the actual number `N`. - Regenerated benchmark results in the environment: | Item | Description | | ---- | ----| | Region | us-west-2 (Oregon) | | Instance | r3.xlarge | | AMI | ubuntu/images/hvm-ssd/ubuntu-bionic-18.04-amd64-server-20190722.1 (ami-06f2f779464715dc5) | | Java | OpenJDK 64-Bit Server VM 1.8.0_242 and OpenJDK 64-Bit Server VM 11.0.6+10 | ### Why are the changes needed? The changes are needed to have: - Correct benchmark results - Base line for other perf improvements that can be checked in the same environment. ### Does this PR introduce _any_ user-facing change? No ### How was this patch tested? By running the benchmark and checking its output. Closes #28440 from MaxGekk/SPARK-31527-DateTimeBenchmark-followup. Authored-by: Max Gekk <max.gekk@gmail.com> Signed-off-by: Takeshi Yamamuro <yamamuro@apache.org>
2020-05-03 20:39:50 -04:00
OpenJDK 64-Bit Server VM 11.0.7+10-post-Ubuntu-2ubuntu218.04 on Linux 4.15.0-1063-aws
Intel(R) Xeon(R) CPU E5-2670 v2 @ 2.50GHz
[SPARK-29320][TESTS] Compare `sql/core` module in JDK8/11 (Part 1) ### What changes were proposed in this pull request? This PR regenerates the `sql/core` benchmarks in JDK8/11 to compare the result. In general, we compare the ratio instead of the time. However, in this PR, the average time is compared. This PR should be considered as a rough comparison. **A. EXPECTED CASES(JDK11 is faster in general)** - [x] BloomFilterBenchmark (JDK11 is faster except one case) - [x] BuiltInDataSourceWriteBenchmark (JDK11 is faster at CSV/ORC) - [x] CSVBenchmark (JDK11 is faster except five cases) - [x] ColumnarBatchBenchmark (JDK11 is faster at `boolean`/`string` and some cases in `int`/`array`) - [x] DatasetBenchmark (JDK11 is faster with `string`, but is slower for `long` type) - [x] ExternalAppendOnlyUnsafeRowArrayBenchmark (JDK11 is faster except two cases) - [x] ExtractBenchmark (JDK11 is faster except HOUR/MINUTE/SECOND/MILLISECONDS/MICROSECONDS) - [x] HashedRelationMetricsBenchmark (JDK11 is faster) - [x] JSONBenchmark (JDK11 is much faster except eight cases) - [x] JoinBenchmark (JDK11 is faster except five cases) - [x] OrcNestedSchemaPruningBenchmark (JDK11 is faster in nine cases) - [x] PrimitiveArrayBenchmark (JDK11 is faster) - [x] SortBenchmark (JDK11 is faster except `Arrays.sort` case) - [x] UDFBenchmark (N/A, values are too small) - [x] UnsafeArrayDataBenchmark (JDK11 is faster except one case) - [x] WideTableBenchmark (JDK11 is faster except two cases) **B. CASES WE NEED TO INVESTIGATE MORE LATER** - [x] AggregateBenchmark (JDK11 is slower in general) - [x] CompressionSchemeBenchmark (JDK11 is slower in general except `string`) - [x] DataSourceReadBenchmark (JDK11 is slower in general) - [x] DateTimeBenchmark (JDK11 is slightly slower in general except `parsing`) - [x] MakeDateTimeBenchmark (JDK11 is slower except two cases) - [x] MiscBenchmark (JDK11 is slower except ten cases) - [x] OrcV2NestedSchemaPruningBenchmark (JDK11 is slower) - [x] ParquetNestedSchemaPruningBenchmark (JDK11 is slower except six cases) - [x] RangeBenchmark (JDK11 is slower except one case) `FilterPushdownBenchmark/InExpressionBenchmark/WideSchemaBenchmark` will be compared later because it took long timer. ### Why are the changes needed? According to the result, there are some difference between JDK8/JDK11. This will be a baseline for the future improvement and comparison. Also, as a reproducible environment, the following environment is used. - Instance: `r3.xlarge` - OS: `CentOS Linux release 7.5.1804 (Core)` - JDK: - `OpenJDK Runtime Environment (build 1.8.0_222-b10)` - `OpenJDK Runtime Environment 18.9 (build 11.0.4+11-LTS)` ### Does this PR introduce any user-facing change? No. ### How was this patch tested? This is a test-only PR. We need to run benchmark. Closes #26003 from dongjoon-hyun/SPARK-29320. Authored-by: Dongjoon Hyun <dhyun@apple.com> Signed-off-by: Dongjoon Hyun <dhyun@apple.com>
2019-10-03 11:58:25 -04:00
dayofyear of timestamp: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative
------------------------------------------------------------------------------------------------------------------------
[SPARK-31630][SQL] Fix perf regression by skipping timestamps rebasing after some threshold ### What changes were proposed in this pull request? Skip timestamps rebasing after a global threshold when there is no difference between Julian and Gregorian calendars. This allows to avoid checking hash maps of switch points, and fixes perf regressions in `toJavaTimestamp()` and `fromJavaTimestamp()`. ### Why are the changes needed? The changes fix perf regressions of conversions to/from external type `java.sql.Timestamp`. Before (see the PR's results https://github.com/apache/spark/pull/28440): ``` ================================================================================================ Conversion from/to external types ================================================================================================ OpenJDK 64-Bit Server VM 1.8.0_252-8u252-b09-1~18.04-b09 on Linux 4.15.0-1063-aws Intel(R) Xeon(R) CPU E5-2670 v2 2.50GHz To/from Java's date-time: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative ------------------------------------------------------------------------------------------------------------------------ From java.sql.Timestamp 376 388 10 13.3 75.2 1.1X Collect java.sql.Timestamp 1878 1937 64 2.7 375.6 0.2X ``` After: ``` ================================================================================================ Conversion from/to external types ================================================================================================ OpenJDK 64-Bit Server VM 1.8.0_252-8u252-b09-1~18.04-b09 on Linux 4.15.0-1063-aws Intel(R) Xeon(R) CPU E5-2670 v2 2.50GHz To/from Java's date-time: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative ------------------------------------------------------------------------------------------------------------------------ From java.sql.Timestamp 249 264 24 20.1 49.8 1.7X Collect java.sql.Timestamp 1503 1523 24 3.3 300.5 0.3X ``` Perf improvements in average of: 1. From java.sql.Timestamp is ~ 34% 2. To java.sql.Timestamps is ~16% ### Does this PR introduce _any_ user-facing change? No ### How was this patch tested? By existing test suites `DateTimeUtilsSuite` and `RebaseDateTimeSuite`. Closes #28441 from MaxGekk/opt-rebase-common-threshold. Authored-by: Max Gekk <max.gekk@gmail.com> Signed-off-by: Wenchen Fan <wenchen@databricks.com>
2020-05-05 10:11:53 -04:00
dayofyear of timestamp wholestage off 1308 1309 1 7.6 130.8 1.0X
dayofyear of timestamp wholestage on 1239 1255 15 8.1 123.9 1.1X
[SPARK-29320][TESTS] Compare `sql/core` module in JDK8/11 (Part 1) ### What changes were proposed in this pull request? This PR regenerates the `sql/core` benchmarks in JDK8/11 to compare the result. In general, we compare the ratio instead of the time. However, in this PR, the average time is compared. This PR should be considered as a rough comparison. **A. EXPECTED CASES(JDK11 is faster in general)** - [x] BloomFilterBenchmark (JDK11 is faster except one case) - [x] BuiltInDataSourceWriteBenchmark (JDK11 is faster at CSV/ORC) - [x] CSVBenchmark (JDK11 is faster except five cases) - [x] ColumnarBatchBenchmark (JDK11 is faster at `boolean`/`string` and some cases in `int`/`array`) - [x] DatasetBenchmark (JDK11 is faster with `string`, but is slower for `long` type) - [x] ExternalAppendOnlyUnsafeRowArrayBenchmark (JDK11 is faster except two cases) - [x] ExtractBenchmark (JDK11 is faster except HOUR/MINUTE/SECOND/MILLISECONDS/MICROSECONDS) - [x] HashedRelationMetricsBenchmark (JDK11 is faster) - [x] JSONBenchmark (JDK11 is much faster except eight cases) - [x] JoinBenchmark (JDK11 is faster except five cases) - [x] OrcNestedSchemaPruningBenchmark (JDK11 is faster in nine cases) - [x] PrimitiveArrayBenchmark (JDK11 is faster) - [x] SortBenchmark (JDK11 is faster except `Arrays.sort` case) - [x] UDFBenchmark (N/A, values are too small) - [x] UnsafeArrayDataBenchmark (JDK11 is faster except one case) - [x] WideTableBenchmark (JDK11 is faster except two cases) **B. CASES WE NEED TO INVESTIGATE MORE LATER** - [x] AggregateBenchmark (JDK11 is slower in general) - [x] CompressionSchemeBenchmark (JDK11 is slower in general except `string`) - [x] DataSourceReadBenchmark (JDK11 is slower in general) - [x] DateTimeBenchmark (JDK11 is slightly slower in general except `parsing`) - [x] MakeDateTimeBenchmark (JDK11 is slower except two cases) - [x] MiscBenchmark (JDK11 is slower except ten cases) - [x] OrcV2NestedSchemaPruningBenchmark (JDK11 is slower) - [x] ParquetNestedSchemaPruningBenchmark (JDK11 is slower except six cases) - [x] RangeBenchmark (JDK11 is slower except one case) `FilterPushdownBenchmark/InExpressionBenchmark/WideSchemaBenchmark` will be compared later because it took long timer. ### Why are the changes needed? According to the result, there are some difference between JDK8/JDK11. This will be a baseline for the future improvement and comparison. Also, as a reproducible environment, the following environment is used. - Instance: `r3.xlarge` - OS: `CentOS Linux release 7.5.1804 (Core)` - JDK: - `OpenJDK Runtime Environment (build 1.8.0_222-b10)` - `OpenJDK Runtime Environment 18.9 (build 11.0.4+11-LTS)` ### Does this PR introduce any user-facing change? No. ### How was this patch tested? This is a test-only PR. We need to run benchmark. Closes #26003 from dongjoon-hyun/SPARK-29320. Authored-by: Dongjoon Hyun <dhyun@apple.com> Signed-off-by: Dongjoon Hyun <dhyun@apple.com>
2019-10-03 11:58:25 -04:00
[SPARK-31527][SQL][TESTS][FOLLOWUP] Fix the number of rows in `DateTimeBenchmark` ### What changes were proposed in this pull request? - Changed to the number of rows in benchmark cases from 3 to the actual number `N`. - Regenerated benchmark results in the environment: | Item | Description | | ---- | ----| | Region | us-west-2 (Oregon) | | Instance | r3.xlarge | | AMI | ubuntu/images/hvm-ssd/ubuntu-bionic-18.04-amd64-server-20190722.1 (ami-06f2f779464715dc5) | | Java | OpenJDK 64-Bit Server VM 1.8.0_242 and OpenJDK 64-Bit Server VM 11.0.6+10 | ### Why are the changes needed? The changes are needed to have: - Correct benchmark results - Base line for other perf improvements that can be checked in the same environment. ### Does this PR introduce _any_ user-facing change? No ### How was this patch tested? By running the benchmark and checking its output. Closes #28440 from MaxGekk/SPARK-31527-DateTimeBenchmark-followup. Authored-by: Max Gekk <max.gekk@gmail.com> Signed-off-by: Takeshi Yamamuro <yamamuro@apache.org>
2020-05-03 20:39:50 -04:00
OpenJDK 64-Bit Server VM 11.0.7+10-post-Ubuntu-2ubuntu218.04 on Linux 4.15.0-1063-aws
Intel(R) Xeon(R) CPU E5-2670 v2 @ 2.50GHz
[SPARK-29320][TESTS] Compare `sql/core` module in JDK8/11 (Part 1) ### What changes were proposed in this pull request? This PR regenerates the `sql/core` benchmarks in JDK8/11 to compare the result. In general, we compare the ratio instead of the time. However, in this PR, the average time is compared. This PR should be considered as a rough comparison. **A. EXPECTED CASES(JDK11 is faster in general)** - [x] BloomFilterBenchmark (JDK11 is faster except one case) - [x] BuiltInDataSourceWriteBenchmark (JDK11 is faster at CSV/ORC) - [x] CSVBenchmark (JDK11 is faster except five cases) - [x] ColumnarBatchBenchmark (JDK11 is faster at `boolean`/`string` and some cases in `int`/`array`) - [x] DatasetBenchmark (JDK11 is faster with `string`, but is slower for `long` type) - [x] ExternalAppendOnlyUnsafeRowArrayBenchmark (JDK11 is faster except two cases) - [x] ExtractBenchmark (JDK11 is faster except HOUR/MINUTE/SECOND/MILLISECONDS/MICROSECONDS) - [x] HashedRelationMetricsBenchmark (JDK11 is faster) - [x] JSONBenchmark (JDK11 is much faster except eight cases) - [x] JoinBenchmark (JDK11 is faster except five cases) - [x] OrcNestedSchemaPruningBenchmark (JDK11 is faster in nine cases) - [x] PrimitiveArrayBenchmark (JDK11 is faster) - [x] SortBenchmark (JDK11 is faster except `Arrays.sort` case) - [x] UDFBenchmark (N/A, values are too small) - [x] UnsafeArrayDataBenchmark (JDK11 is faster except one case) - [x] WideTableBenchmark (JDK11 is faster except two cases) **B. CASES WE NEED TO INVESTIGATE MORE LATER** - [x] AggregateBenchmark (JDK11 is slower in general) - [x] CompressionSchemeBenchmark (JDK11 is slower in general except `string`) - [x] DataSourceReadBenchmark (JDK11 is slower in general) - [x] DateTimeBenchmark (JDK11 is slightly slower in general except `parsing`) - [x] MakeDateTimeBenchmark (JDK11 is slower except two cases) - [x] MiscBenchmark (JDK11 is slower except ten cases) - [x] OrcV2NestedSchemaPruningBenchmark (JDK11 is slower) - [x] ParquetNestedSchemaPruningBenchmark (JDK11 is slower except six cases) - [x] RangeBenchmark (JDK11 is slower except one case) `FilterPushdownBenchmark/InExpressionBenchmark/WideSchemaBenchmark` will be compared later because it took long timer. ### Why are the changes needed? According to the result, there are some difference between JDK8/JDK11. This will be a baseline for the future improvement and comparison. Also, as a reproducible environment, the following environment is used. - Instance: `r3.xlarge` - OS: `CentOS Linux release 7.5.1804 (Core)` - JDK: - `OpenJDK Runtime Environment (build 1.8.0_222-b10)` - `OpenJDK Runtime Environment 18.9 (build 11.0.4+11-LTS)` ### Does this PR introduce any user-facing change? No. ### How was this patch tested? This is a test-only PR. We need to run benchmark. Closes #26003 from dongjoon-hyun/SPARK-29320. Authored-by: Dongjoon Hyun <dhyun@apple.com> Signed-off-by: Dongjoon Hyun <dhyun@apple.com>
2019-10-03 11:58:25 -04:00
dayofmonth of timestamp: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative
------------------------------------------------------------------------------------------------------------------------
[SPARK-31630][SQL] Fix perf regression by skipping timestamps rebasing after some threshold ### What changes were proposed in this pull request? Skip timestamps rebasing after a global threshold when there is no difference between Julian and Gregorian calendars. This allows to avoid checking hash maps of switch points, and fixes perf regressions in `toJavaTimestamp()` and `fromJavaTimestamp()`. ### Why are the changes needed? The changes fix perf regressions of conversions to/from external type `java.sql.Timestamp`. Before (see the PR's results https://github.com/apache/spark/pull/28440): ``` ================================================================================================ Conversion from/to external types ================================================================================================ OpenJDK 64-Bit Server VM 1.8.0_252-8u252-b09-1~18.04-b09 on Linux 4.15.0-1063-aws Intel(R) Xeon(R) CPU E5-2670 v2 2.50GHz To/from Java's date-time: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative ------------------------------------------------------------------------------------------------------------------------ From java.sql.Timestamp 376 388 10 13.3 75.2 1.1X Collect java.sql.Timestamp 1878 1937 64 2.7 375.6 0.2X ``` After: ``` ================================================================================================ Conversion from/to external types ================================================================================================ OpenJDK 64-Bit Server VM 1.8.0_252-8u252-b09-1~18.04-b09 on Linux 4.15.0-1063-aws Intel(R) Xeon(R) CPU E5-2670 v2 2.50GHz To/from Java's date-time: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative ------------------------------------------------------------------------------------------------------------------------ From java.sql.Timestamp 249 264 24 20.1 49.8 1.7X Collect java.sql.Timestamp 1503 1523 24 3.3 300.5 0.3X ``` Perf improvements in average of: 1. From java.sql.Timestamp is ~ 34% 2. To java.sql.Timestamps is ~16% ### Does this PR introduce _any_ user-facing change? No ### How was this patch tested? By existing test suites `DateTimeUtilsSuite` and `RebaseDateTimeSuite`. Closes #28441 from MaxGekk/opt-rebase-common-threshold. Authored-by: Max Gekk <max.gekk@gmail.com> Signed-off-by: Wenchen Fan <wenchen@databricks.com>
2020-05-05 10:11:53 -04:00
dayofmonth of timestamp wholestage off 1259 1263 5 7.9 125.9 1.0X
dayofmonth of timestamp wholestage on 1201 1205 5 8.3 120.1 1.0X
[SPARK-29320][TESTS] Compare `sql/core` module in JDK8/11 (Part 1) ### What changes were proposed in this pull request? This PR regenerates the `sql/core` benchmarks in JDK8/11 to compare the result. In general, we compare the ratio instead of the time. However, in this PR, the average time is compared. This PR should be considered as a rough comparison. **A. EXPECTED CASES(JDK11 is faster in general)** - [x] BloomFilterBenchmark (JDK11 is faster except one case) - [x] BuiltInDataSourceWriteBenchmark (JDK11 is faster at CSV/ORC) - [x] CSVBenchmark (JDK11 is faster except five cases) - [x] ColumnarBatchBenchmark (JDK11 is faster at `boolean`/`string` and some cases in `int`/`array`) - [x] DatasetBenchmark (JDK11 is faster with `string`, but is slower for `long` type) - [x] ExternalAppendOnlyUnsafeRowArrayBenchmark (JDK11 is faster except two cases) - [x] ExtractBenchmark (JDK11 is faster except HOUR/MINUTE/SECOND/MILLISECONDS/MICROSECONDS) - [x] HashedRelationMetricsBenchmark (JDK11 is faster) - [x] JSONBenchmark (JDK11 is much faster except eight cases) - [x] JoinBenchmark (JDK11 is faster except five cases) - [x] OrcNestedSchemaPruningBenchmark (JDK11 is faster in nine cases) - [x] PrimitiveArrayBenchmark (JDK11 is faster) - [x] SortBenchmark (JDK11 is faster except `Arrays.sort` case) - [x] UDFBenchmark (N/A, values are too small) - [x] UnsafeArrayDataBenchmark (JDK11 is faster except one case) - [x] WideTableBenchmark (JDK11 is faster except two cases) **B. CASES WE NEED TO INVESTIGATE MORE LATER** - [x] AggregateBenchmark (JDK11 is slower in general) - [x] CompressionSchemeBenchmark (JDK11 is slower in general except `string`) - [x] DataSourceReadBenchmark (JDK11 is slower in general) - [x] DateTimeBenchmark (JDK11 is slightly slower in general except `parsing`) - [x] MakeDateTimeBenchmark (JDK11 is slower except two cases) - [x] MiscBenchmark (JDK11 is slower except ten cases) - [x] OrcV2NestedSchemaPruningBenchmark (JDK11 is slower) - [x] ParquetNestedSchemaPruningBenchmark (JDK11 is slower except six cases) - [x] RangeBenchmark (JDK11 is slower except one case) `FilterPushdownBenchmark/InExpressionBenchmark/WideSchemaBenchmark` will be compared later because it took long timer. ### Why are the changes needed? According to the result, there are some difference between JDK8/JDK11. This will be a baseline for the future improvement and comparison. Also, as a reproducible environment, the following environment is used. - Instance: `r3.xlarge` - OS: `CentOS Linux release 7.5.1804 (Core)` - JDK: - `OpenJDK Runtime Environment (build 1.8.0_222-b10)` - `OpenJDK Runtime Environment 18.9 (build 11.0.4+11-LTS)` ### Does this PR introduce any user-facing change? No. ### How was this patch tested? This is a test-only PR. We need to run benchmark. Closes #26003 from dongjoon-hyun/SPARK-29320. Authored-by: Dongjoon Hyun <dhyun@apple.com> Signed-off-by: Dongjoon Hyun <dhyun@apple.com>
2019-10-03 11:58:25 -04:00
[SPARK-31527][SQL][TESTS][FOLLOWUP] Fix the number of rows in `DateTimeBenchmark` ### What changes were proposed in this pull request? - Changed to the number of rows in benchmark cases from 3 to the actual number `N`. - Regenerated benchmark results in the environment: | Item | Description | | ---- | ----| | Region | us-west-2 (Oregon) | | Instance | r3.xlarge | | AMI | ubuntu/images/hvm-ssd/ubuntu-bionic-18.04-amd64-server-20190722.1 (ami-06f2f779464715dc5) | | Java | OpenJDK 64-Bit Server VM 1.8.0_242 and OpenJDK 64-Bit Server VM 11.0.6+10 | ### Why are the changes needed? The changes are needed to have: - Correct benchmark results - Base line for other perf improvements that can be checked in the same environment. ### Does this PR introduce _any_ user-facing change? No ### How was this patch tested? By running the benchmark and checking its output. Closes #28440 from MaxGekk/SPARK-31527-DateTimeBenchmark-followup. Authored-by: Max Gekk <max.gekk@gmail.com> Signed-off-by: Takeshi Yamamuro <yamamuro@apache.org>
2020-05-03 20:39:50 -04:00
OpenJDK 64-Bit Server VM 11.0.7+10-post-Ubuntu-2ubuntu218.04 on Linux 4.15.0-1063-aws
Intel(R) Xeon(R) CPU E5-2670 v2 @ 2.50GHz
[SPARK-29320][TESTS] Compare `sql/core` module in JDK8/11 (Part 1) ### What changes were proposed in this pull request? This PR regenerates the `sql/core` benchmarks in JDK8/11 to compare the result. In general, we compare the ratio instead of the time. However, in this PR, the average time is compared. This PR should be considered as a rough comparison. **A. EXPECTED CASES(JDK11 is faster in general)** - [x] BloomFilterBenchmark (JDK11 is faster except one case) - [x] BuiltInDataSourceWriteBenchmark (JDK11 is faster at CSV/ORC) - [x] CSVBenchmark (JDK11 is faster except five cases) - [x] ColumnarBatchBenchmark (JDK11 is faster at `boolean`/`string` and some cases in `int`/`array`) - [x] DatasetBenchmark (JDK11 is faster with `string`, but is slower for `long` type) - [x] ExternalAppendOnlyUnsafeRowArrayBenchmark (JDK11 is faster except two cases) - [x] ExtractBenchmark (JDK11 is faster except HOUR/MINUTE/SECOND/MILLISECONDS/MICROSECONDS) - [x] HashedRelationMetricsBenchmark (JDK11 is faster) - [x] JSONBenchmark (JDK11 is much faster except eight cases) - [x] JoinBenchmark (JDK11 is faster except five cases) - [x] OrcNestedSchemaPruningBenchmark (JDK11 is faster in nine cases) - [x] PrimitiveArrayBenchmark (JDK11 is faster) - [x] SortBenchmark (JDK11 is faster except `Arrays.sort` case) - [x] UDFBenchmark (N/A, values are too small) - [x] UnsafeArrayDataBenchmark (JDK11 is faster except one case) - [x] WideTableBenchmark (JDK11 is faster except two cases) **B. CASES WE NEED TO INVESTIGATE MORE LATER** - [x] AggregateBenchmark (JDK11 is slower in general) - [x] CompressionSchemeBenchmark (JDK11 is slower in general except `string`) - [x] DataSourceReadBenchmark (JDK11 is slower in general) - [x] DateTimeBenchmark (JDK11 is slightly slower in general except `parsing`) - [x] MakeDateTimeBenchmark (JDK11 is slower except two cases) - [x] MiscBenchmark (JDK11 is slower except ten cases) - [x] OrcV2NestedSchemaPruningBenchmark (JDK11 is slower) - [x] ParquetNestedSchemaPruningBenchmark (JDK11 is slower except six cases) - [x] RangeBenchmark (JDK11 is slower except one case) `FilterPushdownBenchmark/InExpressionBenchmark/WideSchemaBenchmark` will be compared later because it took long timer. ### Why are the changes needed? According to the result, there are some difference between JDK8/JDK11. This will be a baseline for the future improvement and comparison. Also, as a reproducible environment, the following environment is used. - Instance: `r3.xlarge` - OS: `CentOS Linux release 7.5.1804 (Core)` - JDK: - `OpenJDK Runtime Environment (build 1.8.0_222-b10)` - `OpenJDK Runtime Environment 18.9 (build 11.0.4+11-LTS)` ### Does this PR introduce any user-facing change? No. ### How was this patch tested? This is a test-only PR. We need to run benchmark. Closes #26003 from dongjoon-hyun/SPARK-29320. Authored-by: Dongjoon Hyun <dhyun@apple.com> Signed-off-by: Dongjoon Hyun <dhyun@apple.com>
2019-10-03 11:58:25 -04:00
dayofweek of timestamp: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative
------------------------------------------------------------------------------------------------------------------------
[SPARK-31630][SQL] Fix perf regression by skipping timestamps rebasing after some threshold ### What changes were proposed in this pull request? Skip timestamps rebasing after a global threshold when there is no difference between Julian and Gregorian calendars. This allows to avoid checking hash maps of switch points, and fixes perf regressions in `toJavaTimestamp()` and `fromJavaTimestamp()`. ### Why are the changes needed? The changes fix perf regressions of conversions to/from external type `java.sql.Timestamp`. Before (see the PR's results https://github.com/apache/spark/pull/28440): ``` ================================================================================================ Conversion from/to external types ================================================================================================ OpenJDK 64-Bit Server VM 1.8.0_252-8u252-b09-1~18.04-b09 on Linux 4.15.0-1063-aws Intel(R) Xeon(R) CPU E5-2670 v2 2.50GHz To/from Java's date-time: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative ------------------------------------------------------------------------------------------------------------------------ From java.sql.Timestamp 376 388 10 13.3 75.2 1.1X Collect java.sql.Timestamp 1878 1937 64 2.7 375.6 0.2X ``` After: ``` ================================================================================================ Conversion from/to external types ================================================================================================ OpenJDK 64-Bit Server VM 1.8.0_252-8u252-b09-1~18.04-b09 on Linux 4.15.0-1063-aws Intel(R) Xeon(R) CPU E5-2670 v2 2.50GHz To/from Java's date-time: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative ------------------------------------------------------------------------------------------------------------------------ From java.sql.Timestamp 249 264 24 20.1 49.8 1.7X Collect java.sql.Timestamp 1503 1523 24 3.3 300.5 0.3X ``` Perf improvements in average of: 1. From java.sql.Timestamp is ~ 34% 2. To java.sql.Timestamps is ~16% ### Does this PR introduce _any_ user-facing change? No ### How was this patch tested? By existing test suites `DateTimeUtilsSuite` and `RebaseDateTimeSuite`. Closes #28441 from MaxGekk/opt-rebase-common-threshold. Authored-by: Max Gekk <max.gekk@gmail.com> Signed-off-by: Wenchen Fan <wenchen@databricks.com>
2020-05-05 10:11:53 -04:00
dayofweek of timestamp wholestage off 1406 1410 6 7.1 140.6 1.0X
dayofweek of timestamp wholestage on 1387 1402 15 7.2 138.7 1.0X
[SPARK-29320][TESTS] Compare `sql/core` module in JDK8/11 (Part 1) ### What changes were proposed in this pull request? This PR regenerates the `sql/core` benchmarks in JDK8/11 to compare the result. In general, we compare the ratio instead of the time. However, in this PR, the average time is compared. This PR should be considered as a rough comparison. **A. EXPECTED CASES(JDK11 is faster in general)** - [x] BloomFilterBenchmark (JDK11 is faster except one case) - [x] BuiltInDataSourceWriteBenchmark (JDK11 is faster at CSV/ORC) - [x] CSVBenchmark (JDK11 is faster except five cases) - [x] ColumnarBatchBenchmark (JDK11 is faster at `boolean`/`string` and some cases in `int`/`array`) - [x] DatasetBenchmark (JDK11 is faster with `string`, but is slower for `long` type) - [x] ExternalAppendOnlyUnsafeRowArrayBenchmark (JDK11 is faster except two cases) - [x] ExtractBenchmark (JDK11 is faster except HOUR/MINUTE/SECOND/MILLISECONDS/MICROSECONDS) - [x] HashedRelationMetricsBenchmark (JDK11 is faster) - [x] JSONBenchmark (JDK11 is much faster except eight cases) - [x] JoinBenchmark (JDK11 is faster except five cases) - [x] OrcNestedSchemaPruningBenchmark (JDK11 is faster in nine cases) - [x] PrimitiveArrayBenchmark (JDK11 is faster) - [x] SortBenchmark (JDK11 is faster except `Arrays.sort` case) - [x] UDFBenchmark (N/A, values are too small) - [x] UnsafeArrayDataBenchmark (JDK11 is faster except one case) - [x] WideTableBenchmark (JDK11 is faster except two cases) **B. CASES WE NEED TO INVESTIGATE MORE LATER** - [x] AggregateBenchmark (JDK11 is slower in general) - [x] CompressionSchemeBenchmark (JDK11 is slower in general except `string`) - [x] DataSourceReadBenchmark (JDK11 is slower in general) - [x] DateTimeBenchmark (JDK11 is slightly slower in general except `parsing`) - [x] MakeDateTimeBenchmark (JDK11 is slower except two cases) - [x] MiscBenchmark (JDK11 is slower except ten cases) - [x] OrcV2NestedSchemaPruningBenchmark (JDK11 is slower) - [x] ParquetNestedSchemaPruningBenchmark (JDK11 is slower except six cases) - [x] RangeBenchmark (JDK11 is slower except one case) `FilterPushdownBenchmark/InExpressionBenchmark/WideSchemaBenchmark` will be compared later because it took long timer. ### Why are the changes needed? According to the result, there are some difference between JDK8/JDK11. This will be a baseline for the future improvement and comparison. Also, as a reproducible environment, the following environment is used. - Instance: `r3.xlarge` - OS: `CentOS Linux release 7.5.1804 (Core)` - JDK: - `OpenJDK Runtime Environment (build 1.8.0_222-b10)` - `OpenJDK Runtime Environment 18.9 (build 11.0.4+11-LTS)` ### Does this PR introduce any user-facing change? No. ### How was this patch tested? This is a test-only PR. We need to run benchmark. Closes #26003 from dongjoon-hyun/SPARK-29320. Authored-by: Dongjoon Hyun <dhyun@apple.com> Signed-off-by: Dongjoon Hyun <dhyun@apple.com>
2019-10-03 11:58:25 -04:00
[SPARK-31527][SQL][TESTS][FOLLOWUP] Fix the number of rows in `DateTimeBenchmark` ### What changes were proposed in this pull request? - Changed to the number of rows in benchmark cases from 3 to the actual number `N`. - Regenerated benchmark results in the environment: | Item | Description | | ---- | ----| | Region | us-west-2 (Oregon) | | Instance | r3.xlarge | | AMI | ubuntu/images/hvm-ssd/ubuntu-bionic-18.04-amd64-server-20190722.1 (ami-06f2f779464715dc5) | | Java | OpenJDK 64-Bit Server VM 1.8.0_242 and OpenJDK 64-Bit Server VM 11.0.6+10 | ### Why are the changes needed? The changes are needed to have: - Correct benchmark results - Base line for other perf improvements that can be checked in the same environment. ### Does this PR introduce _any_ user-facing change? No ### How was this patch tested? By running the benchmark and checking its output. Closes #28440 from MaxGekk/SPARK-31527-DateTimeBenchmark-followup. Authored-by: Max Gekk <max.gekk@gmail.com> Signed-off-by: Takeshi Yamamuro <yamamuro@apache.org>
2020-05-03 20:39:50 -04:00
OpenJDK 64-Bit Server VM 11.0.7+10-post-Ubuntu-2ubuntu218.04 on Linux 4.15.0-1063-aws
Intel(R) Xeon(R) CPU E5-2670 v2 @ 2.50GHz
[SPARK-29320][TESTS] Compare `sql/core` module in JDK8/11 (Part 1) ### What changes were proposed in this pull request? This PR regenerates the `sql/core` benchmarks in JDK8/11 to compare the result. In general, we compare the ratio instead of the time. However, in this PR, the average time is compared. This PR should be considered as a rough comparison. **A. EXPECTED CASES(JDK11 is faster in general)** - [x] BloomFilterBenchmark (JDK11 is faster except one case) - [x] BuiltInDataSourceWriteBenchmark (JDK11 is faster at CSV/ORC) - [x] CSVBenchmark (JDK11 is faster except five cases) - [x] ColumnarBatchBenchmark (JDK11 is faster at `boolean`/`string` and some cases in `int`/`array`) - [x] DatasetBenchmark (JDK11 is faster with `string`, but is slower for `long` type) - [x] ExternalAppendOnlyUnsafeRowArrayBenchmark (JDK11 is faster except two cases) - [x] ExtractBenchmark (JDK11 is faster except HOUR/MINUTE/SECOND/MILLISECONDS/MICROSECONDS) - [x] HashedRelationMetricsBenchmark (JDK11 is faster) - [x] JSONBenchmark (JDK11 is much faster except eight cases) - [x] JoinBenchmark (JDK11 is faster except five cases) - [x] OrcNestedSchemaPruningBenchmark (JDK11 is faster in nine cases) - [x] PrimitiveArrayBenchmark (JDK11 is faster) - [x] SortBenchmark (JDK11 is faster except `Arrays.sort` case) - [x] UDFBenchmark (N/A, values are too small) - [x] UnsafeArrayDataBenchmark (JDK11 is faster except one case) - [x] WideTableBenchmark (JDK11 is faster except two cases) **B. CASES WE NEED TO INVESTIGATE MORE LATER** - [x] AggregateBenchmark (JDK11 is slower in general) - [x] CompressionSchemeBenchmark (JDK11 is slower in general except `string`) - [x] DataSourceReadBenchmark (JDK11 is slower in general) - [x] DateTimeBenchmark (JDK11 is slightly slower in general except `parsing`) - [x] MakeDateTimeBenchmark (JDK11 is slower except two cases) - [x] MiscBenchmark (JDK11 is slower except ten cases) - [x] OrcV2NestedSchemaPruningBenchmark (JDK11 is slower) - [x] ParquetNestedSchemaPruningBenchmark (JDK11 is slower except six cases) - [x] RangeBenchmark (JDK11 is slower except one case) `FilterPushdownBenchmark/InExpressionBenchmark/WideSchemaBenchmark` will be compared later because it took long timer. ### Why are the changes needed? According to the result, there are some difference between JDK8/JDK11. This will be a baseline for the future improvement and comparison. Also, as a reproducible environment, the following environment is used. - Instance: `r3.xlarge` - OS: `CentOS Linux release 7.5.1804 (Core)` - JDK: - `OpenJDK Runtime Environment (build 1.8.0_222-b10)` - `OpenJDK Runtime Environment 18.9 (build 11.0.4+11-LTS)` ### Does this PR introduce any user-facing change? No. ### How was this patch tested? This is a test-only PR. We need to run benchmark. Closes #26003 from dongjoon-hyun/SPARK-29320. Authored-by: Dongjoon Hyun <dhyun@apple.com> Signed-off-by: Dongjoon Hyun <dhyun@apple.com>
2019-10-03 11:58:25 -04:00
weekday of timestamp: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative
------------------------------------------------------------------------------------------------------------------------
[SPARK-31630][SQL] Fix perf regression by skipping timestamps rebasing after some threshold ### What changes were proposed in this pull request? Skip timestamps rebasing after a global threshold when there is no difference between Julian and Gregorian calendars. This allows to avoid checking hash maps of switch points, and fixes perf regressions in `toJavaTimestamp()` and `fromJavaTimestamp()`. ### Why are the changes needed? The changes fix perf regressions of conversions to/from external type `java.sql.Timestamp`. Before (see the PR's results https://github.com/apache/spark/pull/28440): ``` ================================================================================================ Conversion from/to external types ================================================================================================ OpenJDK 64-Bit Server VM 1.8.0_252-8u252-b09-1~18.04-b09 on Linux 4.15.0-1063-aws Intel(R) Xeon(R) CPU E5-2670 v2 2.50GHz To/from Java's date-time: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative ------------------------------------------------------------------------------------------------------------------------ From java.sql.Timestamp 376 388 10 13.3 75.2 1.1X Collect java.sql.Timestamp 1878 1937 64 2.7 375.6 0.2X ``` After: ``` ================================================================================================ Conversion from/to external types ================================================================================================ OpenJDK 64-Bit Server VM 1.8.0_252-8u252-b09-1~18.04-b09 on Linux 4.15.0-1063-aws Intel(R) Xeon(R) CPU E5-2670 v2 2.50GHz To/from Java's date-time: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative ------------------------------------------------------------------------------------------------------------------------ From java.sql.Timestamp 249 264 24 20.1 49.8 1.7X Collect java.sql.Timestamp 1503 1523 24 3.3 300.5 0.3X ``` Perf improvements in average of: 1. From java.sql.Timestamp is ~ 34% 2. To java.sql.Timestamps is ~16% ### Does this PR introduce _any_ user-facing change? No ### How was this patch tested? By existing test suites `DateTimeUtilsSuite` and `RebaseDateTimeSuite`. Closes #28441 from MaxGekk/opt-rebase-common-threshold. Authored-by: Max Gekk <max.gekk@gmail.com> Signed-off-by: Wenchen Fan <wenchen@databricks.com>
2020-05-05 10:11:53 -04:00
weekday of timestamp wholestage off 1355 1367 18 7.4 135.5 1.0X
weekday of timestamp wholestage on 1311 1321 10 7.6 131.1 1.0X
[SPARK-29320][TESTS] Compare `sql/core` module in JDK8/11 (Part 1) ### What changes were proposed in this pull request? This PR regenerates the `sql/core` benchmarks in JDK8/11 to compare the result. In general, we compare the ratio instead of the time. However, in this PR, the average time is compared. This PR should be considered as a rough comparison. **A. EXPECTED CASES(JDK11 is faster in general)** - [x] BloomFilterBenchmark (JDK11 is faster except one case) - [x] BuiltInDataSourceWriteBenchmark (JDK11 is faster at CSV/ORC) - [x] CSVBenchmark (JDK11 is faster except five cases) - [x] ColumnarBatchBenchmark (JDK11 is faster at `boolean`/`string` and some cases in `int`/`array`) - [x] DatasetBenchmark (JDK11 is faster with `string`, but is slower for `long` type) - [x] ExternalAppendOnlyUnsafeRowArrayBenchmark (JDK11 is faster except two cases) - [x] ExtractBenchmark (JDK11 is faster except HOUR/MINUTE/SECOND/MILLISECONDS/MICROSECONDS) - [x] HashedRelationMetricsBenchmark (JDK11 is faster) - [x] JSONBenchmark (JDK11 is much faster except eight cases) - [x] JoinBenchmark (JDK11 is faster except five cases) - [x] OrcNestedSchemaPruningBenchmark (JDK11 is faster in nine cases) - [x] PrimitiveArrayBenchmark (JDK11 is faster) - [x] SortBenchmark (JDK11 is faster except `Arrays.sort` case) - [x] UDFBenchmark (N/A, values are too small) - [x] UnsafeArrayDataBenchmark (JDK11 is faster except one case) - [x] WideTableBenchmark (JDK11 is faster except two cases) **B. CASES WE NEED TO INVESTIGATE MORE LATER** - [x] AggregateBenchmark (JDK11 is slower in general) - [x] CompressionSchemeBenchmark (JDK11 is slower in general except `string`) - [x] DataSourceReadBenchmark (JDK11 is slower in general) - [x] DateTimeBenchmark (JDK11 is slightly slower in general except `parsing`) - [x] MakeDateTimeBenchmark (JDK11 is slower except two cases) - [x] MiscBenchmark (JDK11 is slower except ten cases) - [x] OrcV2NestedSchemaPruningBenchmark (JDK11 is slower) - [x] ParquetNestedSchemaPruningBenchmark (JDK11 is slower except six cases) - [x] RangeBenchmark (JDK11 is slower except one case) `FilterPushdownBenchmark/InExpressionBenchmark/WideSchemaBenchmark` will be compared later because it took long timer. ### Why are the changes needed? According to the result, there are some difference between JDK8/JDK11. This will be a baseline for the future improvement and comparison. Also, as a reproducible environment, the following environment is used. - Instance: `r3.xlarge` - OS: `CentOS Linux release 7.5.1804 (Core)` - JDK: - `OpenJDK Runtime Environment (build 1.8.0_222-b10)` - `OpenJDK Runtime Environment 18.9 (build 11.0.4+11-LTS)` ### Does this PR introduce any user-facing change? No. ### How was this patch tested? This is a test-only PR. We need to run benchmark. Closes #26003 from dongjoon-hyun/SPARK-29320. Authored-by: Dongjoon Hyun <dhyun@apple.com> Signed-off-by: Dongjoon Hyun <dhyun@apple.com>
2019-10-03 11:58:25 -04:00
[SPARK-31527][SQL][TESTS][FOLLOWUP] Fix the number of rows in `DateTimeBenchmark` ### What changes were proposed in this pull request? - Changed to the number of rows in benchmark cases from 3 to the actual number `N`. - Regenerated benchmark results in the environment: | Item | Description | | ---- | ----| | Region | us-west-2 (Oregon) | | Instance | r3.xlarge | | AMI | ubuntu/images/hvm-ssd/ubuntu-bionic-18.04-amd64-server-20190722.1 (ami-06f2f779464715dc5) | | Java | OpenJDK 64-Bit Server VM 1.8.0_242 and OpenJDK 64-Bit Server VM 11.0.6+10 | ### Why are the changes needed? The changes are needed to have: - Correct benchmark results - Base line for other perf improvements that can be checked in the same environment. ### Does this PR introduce _any_ user-facing change? No ### How was this patch tested? By running the benchmark and checking its output. Closes #28440 from MaxGekk/SPARK-31527-DateTimeBenchmark-followup. Authored-by: Max Gekk <max.gekk@gmail.com> Signed-off-by: Takeshi Yamamuro <yamamuro@apache.org>
2020-05-03 20:39:50 -04:00
OpenJDK 64-Bit Server VM 11.0.7+10-post-Ubuntu-2ubuntu218.04 on Linux 4.15.0-1063-aws
Intel(R) Xeon(R) CPU E5-2670 v2 @ 2.50GHz
[SPARK-29320][TESTS] Compare `sql/core` module in JDK8/11 (Part 1) ### What changes were proposed in this pull request? This PR regenerates the `sql/core` benchmarks in JDK8/11 to compare the result. In general, we compare the ratio instead of the time. However, in this PR, the average time is compared. This PR should be considered as a rough comparison. **A. EXPECTED CASES(JDK11 is faster in general)** - [x] BloomFilterBenchmark (JDK11 is faster except one case) - [x] BuiltInDataSourceWriteBenchmark (JDK11 is faster at CSV/ORC) - [x] CSVBenchmark (JDK11 is faster except five cases) - [x] ColumnarBatchBenchmark (JDK11 is faster at `boolean`/`string` and some cases in `int`/`array`) - [x] DatasetBenchmark (JDK11 is faster with `string`, but is slower for `long` type) - [x] ExternalAppendOnlyUnsafeRowArrayBenchmark (JDK11 is faster except two cases) - [x] ExtractBenchmark (JDK11 is faster except HOUR/MINUTE/SECOND/MILLISECONDS/MICROSECONDS) - [x] HashedRelationMetricsBenchmark (JDK11 is faster) - [x] JSONBenchmark (JDK11 is much faster except eight cases) - [x] JoinBenchmark (JDK11 is faster except five cases) - [x] OrcNestedSchemaPruningBenchmark (JDK11 is faster in nine cases) - [x] PrimitiveArrayBenchmark (JDK11 is faster) - [x] SortBenchmark (JDK11 is faster except `Arrays.sort` case) - [x] UDFBenchmark (N/A, values are too small) - [x] UnsafeArrayDataBenchmark (JDK11 is faster except one case) - [x] WideTableBenchmark (JDK11 is faster except two cases) **B. CASES WE NEED TO INVESTIGATE MORE LATER** - [x] AggregateBenchmark (JDK11 is slower in general) - [x] CompressionSchemeBenchmark (JDK11 is slower in general except `string`) - [x] DataSourceReadBenchmark (JDK11 is slower in general) - [x] DateTimeBenchmark (JDK11 is slightly slower in general except `parsing`) - [x] MakeDateTimeBenchmark (JDK11 is slower except two cases) - [x] MiscBenchmark (JDK11 is slower except ten cases) - [x] OrcV2NestedSchemaPruningBenchmark (JDK11 is slower) - [x] ParquetNestedSchemaPruningBenchmark (JDK11 is slower except six cases) - [x] RangeBenchmark (JDK11 is slower except one case) `FilterPushdownBenchmark/InExpressionBenchmark/WideSchemaBenchmark` will be compared later because it took long timer. ### Why are the changes needed? According to the result, there are some difference between JDK8/JDK11. This will be a baseline for the future improvement and comparison. Also, as a reproducible environment, the following environment is used. - Instance: `r3.xlarge` - OS: `CentOS Linux release 7.5.1804 (Core)` - JDK: - `OpenJDK Runtime Environment (build 1.8.0_222-b10)` - `OpenJDK Runtime Environment 18.9 (build 11.0.4+11-LTS)` ### Does this PR introduce any user-facing change? No. ### How was this patch tested? This is a test-only PR. We need to run benchmark. Closes #26003 from dongjoon-hyun/SPARK-29320. Authored-by: Dongjoon Hyun <dhyun@apple.com> Signed-off-by: Dongjoon Hyun <dhyun@apple.com>
2019-10-03 11:58:25 -04:00
hour of timestamp: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative
------------------------------------------------------------------------------------------------------------------------
[SPARK-31630][SQL] Fix perf regression by skipping timestamps rebasing after some threshold ### What changes were proposed in this pull request? Skip timestamps rebasing after a global threshold when there is no difference between Julian and Gregorian calendars. This allows to avoid checking hash maps of switch points, and fixes perf regressions in `toJavaTimestamp()` and `fromJavaTimestamp()`. ### Why are the changes needed? The changes fix perf regressions of conversions to/from external type `java.sql.Timestamp`. Before (see the PR's results https://github.com/apache/spark/pull/28440): ``` ================================================================================================ Conversion from/to external types ================================================================================================ OpenJDK 64-Bit Server VM 1.8.0_252-8u252-b09-1~18.04-b09 on Linux 4.15.0-1063-aws Intel(R) Xeon(R) CPU E5-2670 v2 2.50GHz To/from Java's date-time: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative ------------------------------------------------------------------------------------------------------------------------ From java.sql.Timestamp 376 388 10 13.3 75.2 1.1X Collect java.sql.Timestamp 1878 1937 64 2.7 375.6 0.2X ``` After: ``` ================================================================================================ Conversion from/to external types ================================================================================================ OpenJDK 64-Bit Server VM 1.8.0_252-8u252-b09-1~18.04-b09 on Linux 4.15.0-1063-aws Intel(R) Xeon(R) CPU E5-2670 v2 2.50GHz To/from Java's date-time: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative ------------------------------------------------------------------------------------------------------------------------ From java.sql.Timestamp 249 264 24 20.1 49.8 1.7X Collect java.sql.Timestamp 1503 1523 24 3.3 300.5 0.3X ``` Perf improvements in average of: 1. From java.sql.Timestamp is ~ 34% 2. To java.sql.Timestamps is ~16% ### Does this PR introduce _any_ user-facing change? No ### How was this patch tested? By existing test suites `DateTimeUtilsSuite` and `RebaseDateTimeSuite`. Closes #28441 from MaxGekk/opt-rebase-common-threshold. Authored-by: Max Gekk <max.gekk@gmail.com> Signed-off-by: Wenchen Fan <wenchen@databricks.com>
2020-05-05 10:11:53 -04:00
hour of timestamp wholestage off 996 997 2 10.0 99.6 1.0X
hour of timestamp wholestage on 930 936 6 10.7 93.0 1.1X
[SPARK-29320][TESTS] Compare `sql/core` module in JDK8/11 (Part 1) ### What changes were proposed in this pull request? This PR regenerates the `sql/core` benchmarks in JDK8/11 to compare the result. In general, we compare the ratio instead of the time. However, in this PR, the average time is compared. This PR should be considered as a rough comparison. **A. EXPECTED CASES(JDK11 is faster in general)** - [x] BloomFilterBenchmark (JDK11 is faster except one case) - [x] BuiltInDataSourceWriteBenchmark (JDK11 is faster at CSV/ORC) - [x] CSVBenchmark (JDK11 is faster except five cases) - [x] ColumnarBatchBenchmark (JDK11 is faster at `boolean`/`string` and some cases in `int`/`array`) - [x] DatasetBenchmark (JDK11 is faster with `string`, but is slower for `long` type) - [x] ExternalAppendOnlyUnsafeRowArrayBenchmark (JDK11 is faster except two cases) - [x] ExtractBenchmark (JDK11 is faster except HOUR/MINUTE/SECOND/MILLISECONDS/MICROSECONDS) - [x] HashedRelationMetricsBenchmark (JDK11 is faster) - [x] JSONBenchmark (JDK11 is much faster except eight cases) - [x] JoinBenchmark (JDK11 is faster except five cases) - [x] OrcNestedSchemaPruningBenchmark (JDK11 is faster in nine cases) - [x] PrimitiveArrayBenchmark (JDK11 is faster) - [x] SortBenchmark (JDK11 is faster except `Arrays.sort` case) - [x] UDFBenchmark (N/A, values are too small) - [x] UnsafeArrayDataBenchmark (JDK11 is faster except one case) - [x] WideTableBenchmark (JDK11 is faster except two cases) **B. CASES WE NEED TO INVESTIGATE MORE LATER** - [x] AggregateBenchmark (JDK11 is slower in general) - [x] CompressionSchemeBenchmark (JDK11 is slower in general except `string`) - [x] DataSourceReadBenchmark (JDK11 is slower in general) - [x] DateTimeBenchmark (JDK11 is slightly slower in general except `parsing`) - [x] MakeDateTimeBenchmark (JDK11 is slower except two cases) - [x] MiscBenchmark (JDK11 is slower except ten cases) - [x] OrcV2NestedSchemaPruningBenchmark (JDK11 is slower) - [x] ParquetNestedSchemaPruningBenchmark (JDK11 is slower except six cases) - [x] RangeBenchmark (JDK11 is slower except one case) `FilterPushdownBenchmark/InExpressionBenchmark/WideSchemaBenchmark` will be compared later because it took long timer. ### Why are the changes needed? According to the result, there are some difference between JDK8/JDK11. This will be a baseline for the future improvement and comparison. Also, as a reproducible environment, the following environment is used. - Instance: `r3.xlarge` - OS: `CentOS Linux release 7.5.1804 (Core)` - JDK: - `OpenJDK Runtime Environment (build 1.8.0_222-b10)` - `OpenJDK Runtime Environment 18.9 (build 11.0.4+11-LTS)` ### Does this PR introduce any user-facing change? No. ### How was this patch tested? This is a test-only PR. We need to run benchmark. Closes #26003 from dongjoon-hyun/SPARK-29320. Authored-by: Dongjoon Hyun <dhyun@apple.com> Signed-off-by: Dongjoon Hyun <dhyun@apple.com>
2019-10-03 11:58:25 -04:00
[SPARK-31527][SQL][TESTS][FOLLOWUP] Fix the number of rows in `DateTimeBenchmark` ### What changes were proposed in this pull request? - Changed to the number of rows in benchmark cases from 3 to the actual number `N`. - Regenerated benchmark results in the environment: | Item | Description | | ---- | ----| | Region | us-west-2 (Oregon) | | Instance | r3.xlarge | | AMI | ubuntu/images/hvm-ssd/ubuntu-bionic-18.04-amd64-server-20190722.1 (ami-06f2f779464715dc5) | | Java | OpenJDK 64-Bit Server VM 1.8.0_242 and OpenJDK 64-Bit Server VM 11.0.6+10 | ### Why are the changes needed? The changes are needed to have: - Correct benchmark results - Base line for other perf improvements that can be checked in the same environment. ### Does this PR introduce _any_ user-facing change? No ### How was this patch tested? By running the benchmark and checking its output. Closes #28440 from MaxGekk/SPARK-31527-DateTimeBenchmark-followup. Authored-by: Max Gekk <max.gekk@gmail.com> Signed-off-by: Takeshi Yamamuro <yamamuro@apache.org>
2020-05-03 20:39:50 -04:00
OpenJDK 64-Bit Server VM 11.0.7+10-post-Ubuntu-2ubuntu218.04 on Linux 4.15.0-1063-aws
Intel(R) Xeon(R) CPU E5-2670 v2 @ 2.50GHz
[SPARK-29320][TESTS] Compare `sql/core` module in JDK8/11 (Part 1) ### What changes were proposed in this pull request? This PR regenerates the `sql/core` benchmarks in JDK8/11 to compare the result. In general, we compare the ratio instead of the time. However, in this PR, the average time is compared. This PR should be considered as a rough comparison. **A. EXPECTED CASES(JDK11 is faster in general)** - [x] BloomFilterBenchmark (JDK11 is faster except one case) - [x] BuiltInDataSourceWriteBenchmark (JDK11 is faster at CSV/ORC) - [x] CSVBenchmark (JDK11 is faster except five cases) - [x] ColumnarBatchBenchmark (JDK11 is faster at `boolean`/`string` and some cases in `int`/`array`) - [x] DatasetBenchmark (JDK11 is faster with `string`, but is slower for `long` type) - [x] ExternalAppendOnlyUnsafeRowArrayBenchmark (JDK11 is faster except two cases) - [x] ExtractBenchmark (JDK11 is faster except HOUR/MINUTE/SECOND/MILLISECONDS/MICROSECONDS) - [x] HashedRelationMetricsBenchmark (JDK11 is faster) - [x] JSONBenchmark (JDK11 is much faster except eight cases) - [x] JoinBenchmark (JDK11 is faster except five cases) - [x] OrcNestedSchemaPruningBenchmark (JDK11 is faster in nine cases) - [x] PrimitiveArrayBenchmark (JDK11 is faster) - [x] SortBenchmark (JDK11 is faster except `Arrays.sort` case) - [x] UDFBenchmark (N/A, values are too small) - [x] UnsafeArrayDataBenchmark (JDK11 is faster except one case) - [x] WideTableBenchmark (JDK11 is faster except two cases) **B. CASES WE NEED TO INVESTIGATE MORE LATER** - [x] AggregateBenchmark (JDK11 is slower in general) - [x] CompressionSchemeBenchmark (JDK11 is slower in general except `string`) - [x] DataSourceReadBenchmark (JDK11 is slower in general) - [x] DateTimeBenchmark (JDK11 is slightly slower in general except `parsing`) - [x] MakeDateTimeBenchmark (JDK11 is slower except two cases) - [x] MiscBenchmark (JDK11 is slower except ten cases) - [x] OrcV2NestedSchemaPruningBenchmark (JDK11 is slower) - [x] ParquetNestedSchemaPruningBenchmark (JDK11 is slower except six cases) - [x] RangeBenchmark (JDK11 is slower except one case) `FilterPushdownBenchmark/InExpressionBenchmark/WideSchemaBenchmark` will be compared later because it took long timer. ### Why are the changes needed? According to the result, there are some difference between JDK8/JDK11. This will be a baseline for the future improvement and comparison. Also, as a reproducible environment, the following environment is used. - Instance: `r3.xlarge` - OS: `CentOS Linux release 7.5.1804 (Core)` - JDK: - `OpenJDK Runtime Environment (build 1.8.0_222-b10)` - `OpenJDK Runtime Environment 18.9 (build 11.0.4+11-LTS)` ### Does this PR introduce any user-facing change? No. ### How was this patch tested? This is a test-only PR. We need to run benchmark. Closes #26003 from dongjoon-hyun/SPARK-29320. Authored-by: Dongjoon Hyun <dhyun@apple.com> Signed-off-by: Dongjoon Hyun <dhyun@apple.com>
2019-10-03 11:58:25 -04:00
minute of timestamp: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative
------------------------------------------------------------------------------------------------------------------------
[SPARK-31630][SQL] Fix perf regression by skipping timestamps rebasing after some threshold ### What changes were proposed in this pull request? Skip timestamps rebasing after a global threshold when there is no difference between Julian and Gregorian calendars. This allows to avoid checking hash maps of switch points, and fixes perf regressions in `toJavaTimestamp()` and `fromJavaTimestamp()`. ### Why are the changes needed? The changes fix perf regressions of conversions to/from external type `java.sql.Timestamp`. Before (see the PR's results https://github.com/apache/spark/pull/28440): ``` ================================================================================================ Conversion from/to external types ================================================================================================ OpenJDK 64-Bit Server VM 1.8.0_252-8u252-b09-1~18.04-b09 on Linux 4.15.0-1063-aws Intel(R) Xeon(R) CPU E5-2670 v2 2.50GHz To/from Java's date-time: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative ------------------------------------------------------------------------------------------------------------------------ From java.sql.Timestamp 376 388 10 13.3 75.2 1.1X Collect java.sql.Timestamp 1878 1937 64 2.7 375.6 0.2X ``` After: ``` ================================================================================================ Conversion from/to external types ================================================================================================ OpenJDK 64-Bit Server VM 1.8.0_252-8u252-b09-1~18.04-b09 on Linux 4.15.0-1063-aws Intel(R) Xeon(R) CPU E5-2670 v2 2.50GHz To/from Java's date-time: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative ------------------------------------------------------------------------------------------------------------------------ From java.sql.Timestamp 249 264 24 20.1 49.8 1.7X Collect java.sql.Timestamp 1503 1523 24 3.3 300.5 0.3X ``` Perf improvements in average of: 1. From java.sql.Timestamp is ~ 34% 2. To java.sql.Timestamps is ~16% ### Does this PR introduce _any_ user-facing change? No ### How was this patch tested? By existing test suites `DateTimeUtilsSuite` and `RebaseDateTimeSuite`. Closes #28441 from MaxGekk/opt-rebase-common-threshold. Authored-by: Max Gekk <max.gekk@gmail.com> Signed-off-by: Wenchen Fan <wenchen@databricks.com>
2020-05-05 10:11:53 -04:00
minute of timestamp wholestage off 1005 1012 10 9.9 100.5 1.0X
minute of timestamp wholestage on 949 952 3 10.5 94.9 1.1X
[SPARK-29320][TESTS] Compare `sql/core` module in JDK8/11 (Part 1) ### What changes were proposed in this pull request? This PR regenerates the `sql/core` benchmarks in JDK8/11 to compare the result. In general, we compare the ratio instead of the time. However, in this PR, the average time is compared. This PR should be considered as a rough comparison. **A. EXPECTED CASES(JDK11 is faster in general)** - [x] BloomFilterBenchmark (JDK11 is faster except one case) - [x] BuiltInDataSourceWriteBenchmark (JDK11 is faster at CSV/ORC) - [x] CSVBenchmark (JDK11 is faster except five cases) - [x] ColumnarBatchBenchmark (JDK11 is faster at `boolean`/`string` and some cases in `int`/`array`) - [x] DatasetBenchmark (JDK11 is faster with `string`, but is slower for `long` type) - [x] ExternalAppendOnlyUnsafeRowArrayBenchmark (JDK11 is faster except two cases) - [x] ExtractBenchmark (JDK11 is faster except HOUR/MINUTE/SECOND/MILLISECONDS/MICROSECONDS) - [x] HashedRelationMetricsBenchmark (JDK11 is faster) - [x] JSONBenchmark (JDK11 is much faster except eight cases) - [x] JoinBenchmark (JDK11 is faster except five cases) - [x] OrcNestedSchemaPruningBenchmark (JDK11 is faster in nine cases) - [x] PrimitiveArrayBenchmark (JDK11 is faster) - [x] SortBenchmark (JDK11 is faster except `Arrays.sort` case) - [x] UDFBenchmark (N/A, values are too small) - [x] UnsafeArrayDataBenchmark (JDK11 is faster except one case) - [x] WideTableBenchmark (JDK11 is faster except two cases) **B. CASES WE NEED TO INVESTIGATE MORE LATER** - [x] AggregateBenchmark (JDK11 is slower in general) - [x] CompressionSchemeBenchmark (JDK11 is slower in general except `string`) - [x] DataSourceReadBenchmark (JDK11 is slower in general) - [x] DateTimeBenchmark (JDK11 is slightly slower in general except `parsing`) - [x] MakeDateTimeBenchmark (JDK11 is slower except two cases) - [x] MiscBenchmark (JDK11 is slower except ten cases) - [x] OrcV2NestedSchemaPruningBenchmark (JDK11 is slower) - [x] ParquetNestedSchemaPruningBenchmark (JDK11 is slower except six cases) - [x] RangeBenchmark (JDK11 is slower except one case) `FilterPushdownBenchmark/InExpressionBenchmark/WideSchemaBenchmark` will be compared later because it took long timer. ### Why are the changes needed? According to the result, there are some difference between JDK8/JDK11. This will be a baseline for the future improvement and comparison. Also, as a reproducible environment, the following environment is used. - Instance: `r3.xlarge` - OS: `CentOS Linux release 7.5.1804 (Core)` - JDK: - `OpenJDK Runtime Environment (build 1.8.0_222-b10)` - `OpenJDK Runtime Environment 18.9 (build 11.0.4+11-LTS)` ### Does this PR introduce any user-facing change? No. ### How was this patch tested? This is a test-only PR. We need to run benchmark. Closes #26003 from dongjoon-hyun/SPARK-29320. Authored-by: Dongjoon Hyun <dhyun@apple.com> Signed-off-by: Dongjoon Hyun <dhyun@apple.com>
2019-10-03 11:58:25 -04:00
[SPARK-31527][SQL][TESTS][FOLLOWUP] Fix the number of rows in `DateTimeBenchmark` ### What changes were proposed in this pull request? - Changed to the number of rows in benchmark cases from 3 to the actual number `N`. - Regenerated benchmark results in the environment: | Item | Description | | ---- | ----| | Region | us-west-2 (Oregon) | | Instance | r3.xlarge | | AMI | ubuntu/images/hvm-ssd/ubuntu-bionic-18.04-amd64-server-20190722.1 (ami-06f2f779464715dc5) | | Java | OpenJDK 64-Bit Server VM 1.8.0_242 and OpenJDK 64-Bit Server VM 11.0.6+10 | ### Why are the changes needed? The changes are needed to have: - Correct benchmark results - Base line for other perf improvements that can be checked in the same environment. ### Does this PR introduce _any_ user-facing change? No ### How was this patch tested? By running the benchmark and checking its output. Closes #28440 from MaxGekk/SPARK-31527-DateTimeBenchmark-followup. Authored-by: Max Gekk <max.gekk@gmail.com> Signed-off-by: Takeshi Yamamuro <yamamuro@apache.org>
2020-05-03 20:39:50 -04:00
OpenJDK 64-Bit Server VM 11.0.7+10-post-Ubuntu-2ubuntu218.04 on Linux 4.15.0-1063-aws
Intel(R) Xeon(R) CPU E5-2670 v2 @ 2.50GHz
[SPARK-29320][TESTS] Compare `sql/core` module in JDK8/11 (Part 1) ### What changes were proposed in this pull request? This PR regenerates the `sql/core` benchmarks in JDK8/11 to compare the result. In general, we compare the ratio instead of the time. However, in this PR, the average time is compared. This PR should be considered as a rough comparison. **A. EXPECTED CASES(JDK11 is faster in general)** - [x] BloomFilterBenchmark (JDK11 is faster except one case) - [x] BuiltInDataSourceWriteBenchmark (JDK11 is faster at CSV/ORC) - [x] CSVBenchmark (JDK11 is faster except five cases) - [x] ColumnarBatchBenchmark (JDK11 is faster at `boolean`/`string` and some cases in `int`/`array`) - [x] DatasetBenchmark (JDK11 is faster with `string`, but is slower for `long` type) - [x] ExternalAppendOnlyUnsafeRowArrayBenchmark (JDK11 is faster except two cases) - [x] ExtractBenchmark (JDK11 is faster except HOUR/MINUTE/SECOND/MILLISECONDS/MICROSECONDS) - [x] HashedRelationMetricsBenchmark (JDK11 is faster) - [x] JSONBenchmark (JDK11 is much faster except eight cases) - [x] JoinBenchmark (JDK11 is faster except five cases) - [x] OrcNestedSchemaPruningBenchmark (JDK11 is faster in nine cases) - [x] PrimitiveArrayBenchmark (JDK11 is faster) - [x] SortBenchmark (JDK11 is faster except `Arrays.sort` case) - [x] UDFBenchmark (N/A, values are too small) - [x] UnsafeArrayDataBenchmark (JDK11 is faster except one case) - [x] WideTableBenchmark (JDK11 is faster except two cases) **B. CASES WE NEED TO INVESTIGATE MORE LATER** - [x] AggregateBenchmark (JDK11 is slower in general) - [x] CompressionSchemeBenchmark (JDK11 is slower in general except `string`) - [x] DataSourceReadBenchmark (JDK11 is slower in general) - [x] DateTimeBenchmark (JDK11 is slightly slower in general except `parsing`) - [x] MakeDateTimeBenchmark (JDK11 is slower except two cases) - [x] MiscBenchmark (JDK11 is slower except ten cases) - [x] OrcV2NestedSchemaPruningBenchmark (JDK11 is slower) - [x] ParquetNestedSchemaPruningBenchmark (JDK11 is slower except six cases) - [x] RangeBenchmark (JDK11 is slower except one case) `FilterPushdownBenchmark/InExpressionBenchmark/WideSchemaBenchmark` will be compared later because it took long timer. ### Why are the changes needed? According to the result, there are some difference between JDK8/JDK11. This will be a baseline for the future improvement and comparison. Also, as a reproducible environment, the following environment is used. - Instance: `r3.xlarge` - OS: `CentOS Linux release 7.5.1804 (Core)` - JDK: - `OpenJDK Runtime Environment (build 1.8.0_222-b10)` - `OpenJDK Runtime Environment 18.9 (build 11.0.4+11-LTS)` ### Does this PR introduce any user-facing change? No. ### How was this patch tested? This is a test-only PR. We need to run benchmark. Closes #26003 from dongjoon-hyun/SPARK-29320. Authored-by: Dongjoon Hyun <dhyun@apple.com> Signed-off-by: Dongjoon Hyun <dhyun@apple.com>
2019-10-03 11:58:25 -04:00
second of timestamp: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative
------------------------------------------------------------------------------------------------------------------------
[SPARK-31630][SQL] Fix perf regression by skipping timestamps rebasing after some threshold ### What changes were proposed in this pull request? Skip timestamps rebasing after a global threshold when there is no difference between Julian and Gregorian calendars. This allows to avoid checking hash maps of switch points, and fixes perf regressions in `toJavaTimestamp()` and `fromJavaTimestamp()`. ### Why are the changes needed? The changes fix perf regressions of conversions to/from external type `java.sql.Timestamp`. Before (see the PR's results https://github.com/apache/spark/pull/28440): ``` ================================================================================================ Conversion from/to external types ================================================================================================ OpenJDK 64-Bit Server VM 1.8.0_252-8u252-b09-1~18.04-b09 on Linux 4.15.0-1063-aws Intel(R) Xeon(R) CPU E5-2670 v2 2.50GHz To/from Java's date-time: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative ------------------------------------------------------------------------------------------------------------------------ From java.sql.Timestamp 376 388 10 13.3 75.2 1.1X Collect java.sql.Timestamp 1878 1937 64 2.7 375.6 0.2X ``` After: ``` ================================================================================================ Conversion from/to external types ================================================================================================ OpenJDK 64-Bit Server VM 1.8.0_252-8u252-b09-1~18.04-b09 on Linux 4.15.0-1063-aws Intel(R) Xeon(R) CPU E5-2670 v2 2.50GHz To/from Java's date-time: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative ------------------------------------------------------------------------------------------------------------------------ From java.sql.Timestamp 249 264 24 20.1 49.8 1.7X Collect java.sql.Timestamp 1503 1523 24 3.3 300.5 0.3X ``` Perf improvements in average of: 1. From java.sql.Timestamp is ~ 34% 2. To java.sql.Timestamps is ~16% ### Does this PR introduce _any_ user-facing change? No ### How was this patch tested? By existing test suites `DateTimeUtilsSuite` and `RebaseDateTimeSuite`. Closes #28441 from MaxGekk/opt-rebase-common-threshold. Authored-by: Max Gekk <max.gekk@gmail.com> Signed-off-by: Wenchen Fan <wenchen@databricks.com>
2020-05-05 10:11:53 -04:00
second of timestamp wholestage off 1013 1014 1 9.9 101.3 1.0X
second of timestamp wholestage on 933 934 2 10.7 93.3 1.1X
[SPARK-29320][TESTS] Compare `sql/core` module in JDK8/11 (Part 1) ### What changes were proposed in this pull request? This PR regenerates the `sql/core` benchmarks in JDK8/11 to compare the result. In general, we compare the ratio instead of the time. However, in this PR, the average time is compared. This PR should be considered as a rough comparison. **A. EXPECTED CASES(JDK11 is faster in general)** - [x] BloomFilterBenchmark (JDK11 is faster except one case) - [x] BuiltInDataSourceWriteBenchmark (JDK11 is faster at CSV/ORC) - [x] CSVBenchmark (JDK11 is faster except five cases) - [x] ColumnarBatchBenchmark (JDK11 is faster at `boolean`/`string` and some cases in `int`/`array`) - [x] DatasetBenchmark (JDK11 is faster with `string`, but is slower for `long` type) - [x] ExternalAppendOnlyUnsafeRowArrayBenchmark (JDK11 is faster except two cases) - [x] ExtractBenchmark (JDK11 is faster except HOUR/MINUTE/SECOND/MILLISECONDS/MICROSECONDS) - [x] HashedRelationMetricsBenchmark (JDK11 is faster) - [x] JSONBenchmark (JDK11 is much faster except eight cases) - [x] JoinBenchmark (JDK11 is faster except five cases) - [x] OrcNestedSchemaPruningBenchmark (JDK11 is faster in nine cases) - [x] PrimitiveArrayBenchmark (JDK11 is faster) - [x] SortBenchmark (JDK11 is faster except `Arrays.sort` case) - [x] UDFBenchmark (N/A, values are too small) - [x] UnsafeArrayDataBenchmark (JDK11 is faster except one case) - [x] WideTableBenchmark (JDK11 is faster except two cases) **B. CASES WE NEED TO INVESTIGATE MORE LATER** - [x] AggregateBenchmark (JDK11 is slower in general) - [x] CompressionSchemeBenchmark (JDK11 is slower in general except `string`) - [x] DataSourceReadBenchmark (JDK11 is slower in general) - [x] DateTimeBenchmark (JDK11 is slightly slower in general except `parsing`) - [x] MakeDateTimeBenchmark (JDK11 is slower except two cases) - [x] MiscBenchmark (JDK11 is slower except ten cases) - [x] OrcV2NestedSchemaPruningBenchmark (JDK11 is slower) - [x] ParquetNestedSchemaPruningBenchmark (JDK11 is slower except six cases) - [x] RangeBenchmark (JDK11 is slower except one case) `FilterPushdownBenchmark/InExpressionBenchmark/WideSchemaBenchmark` will be compared later because it took long timer. ### Why are the changes needed? According to the result, there are some difference between JDK8/JDK11. This will be a baseline for the future improvement and comparison. Also, as a reproducible environment, the following environment is used. - Instance: `r3.xlarge` - OS: `CentOS Linux release 7.5.1804 (Core)` - JDK: - `OpenJDK Runtime Environment (build 1.8.0_222-b10)` - `OpenJDK Runtime Environment 18.9 (build 11.0.4+11-LTS)` ### Does this PR introduce any user-facing change? No. ### How was this patch tested? This is a test-only PR. We need to run benchmark. Closes #26003 from dongjoon-hyun/SPARK-29320. Authored-by: Dongjoon Hyun <dhyun@apple.com> Signed-off-by: Dongjoon Hyun <dhyun@apple.com>
2019-10-03 11:58:25 -04:00
================================================================================================
Current date and time
================================================================================================
[SPARK-31527][SQL][TESTS][FOLLOWUP] Fix the number of rows in `DateTimeBenchmark` ### What changes were proposed in this pull request? - Changed to the number of rows in benchmark cases from 3 to the actual number `N`. - Regenerated benchmark results in the environment: | Item | Description | | ---- | ----| | Region | us-west-2 (Oregon) | | Instance | r3.xlarge | | AMI | ubuntu/images/hvm-ssd/ubuntu-bionic-18.04-amd64-server-20190722.1 (ami-06f2f779464715dc5) | | Java | OpenJDK 64-Bit Server VM 1.8.0_242 and OpenJDK 64-Bit Server VM 11.0.6+10 | ### Why are the changes needed? The changes are needed to have: - Correct benchmark results - Base line for other perf improvements that can be checked in the same environment. ### Does this PR introduce _any_ user-facing change? No ### How was this patch tested? By running the benchmark and checking its output. Closes #28440 from MaxGekk/SPARK-31527-DateTimeBenchmark-followup. Authored-by: Max Gekk <max.gekk@gmail.com> Signed-off-by: Takeshi Yamamuro <yamamuro@apache.org>
2020-05-03 20:39:50 -04:00
OpenJDK 64-Bit Server VM 11.0.7+10-post-Ubuntu-2ubuntu218.04 on Linux 4.15.0-1063-aws
Intel(R) Xeon(R) CPU E5-2670 v2 @ 2.50GHz
[SPARK-29320][TESTS] Compare `sql/core` module in JDK8/11 (Part 1) ### What changes were proposed in this pull request? This PR regenerates the `sql/core` benchmarks in JDK8/11 to compare the result. In general, we compare the ratio instead of the time. However, in this PR, the average time is compared. This PR should be considered as a rough comparison. **A. EXPECTED CASES(JDK11 is faster in general)** - [x] BloomFilterBenchmark (JDK11 is faster except one case) - [x] BuiltInDataSourceWriteBenchmark (JDK11 is faster at CSV/ORC) - [x] CSVBenchmark (JDK11 is faster except five cases) - [x] ColumnarBatchBenchmark (JDK11 is faster at `boolean`/`string` and some cases in `int`/`array`) - [x] DatasetBenchmark (JDK11 is faster with `string`, but is slower for `long` type) - [x] ExternalAppendOnlyUnsafeRowArrayBenchmark (JDK11 is faster except two cases) - [x] ExtractBenchmark (JDK11 is faster except HOUR/MINUTE/SECOND/MILLISECONDS/MICROSECONDS) - [x] HashedRelationMetricsBenchmark (JDK11 is faster) - [x] JSONBenchmark (JDK11 is much faster except eight cases) - [x] JoinBenchmark (JDK11 is faster except five cases) - [x] OrcNestedSchemaPruningBenchmark (JDK11 is faster in nine cases) - [x] PrimitiveArrayBenchmark (JDK11 is faster) - [x] SortBenchmark (JDK11 is faster except `Arrays.sort` case) - [x] UDFBenchmark (N/A, values are too small) - [x] UnsafeArrayDataBenchmark (JDK11 is faster except one case) - [x] WideTableBenchmark (JDK11 is faster except two cases) **B. CASES WE NEED TO INVESTIGATE MORE LATER** - [x] AggregateBenchmark (JDK11 is slower in general) - [x] CompressionSchemeBenchmark (JDK11 is slower in general except `string`) - [x] DataSourceReadBenchmark (JDK11 is slower in general) - [x] DateTimeBenchmark (JDK11 is slightly slower in general except `parsing`) - [x] MakeDateTimeBenchmark (JDK11 is slower except two cases) - [x] MiscBenchmark (JDK11 is slower except ten cases) - [x] OrcV2NestedSchemaPruningBenchmark (JDK11 is slower) - [x] ParquetNestedSchemaPruningBenchmark (JDK11 is slower except six cases) - [x] RangeBenchmark (JDK11 is slower except one case) `FilterPushdownBenchmark/InExpressionBenchmark/WideSchemaBenchmark` will be compared later because it took long timer. ### Why are the changes needed? According to the result, there are some difference between JDK8/JDK11. This will be a baseline for the future improvement and comparison. Also, as a reproducible environment, the following environment is used. - Instance: `r3.xlarge` - OS: `CentOS Linux release 7.5.1804 (Core)` - JDK: - `OpenJDK Runtime Environment (build 1.8.0_222-b10)` - `OpenJDK Runtime Environment 18.9 (build 11.0.4+11-LTS)` ### Does this PR introduce any user-facing change? No. ### How was this patch tested? This is a test-only PR. We need to run benchmark. Closes #26003 from dongjoon-hyun/SPARK-29320. Authored-by: Dongjoon Hyun <dhyun@apple.com> Signed-off-by: Dongjoon Hyun <dhyun@apple.com>
2019-10-03 11:58:25 -04:00
current_date: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative
------------------------------------------------------------------------------------------------------------------------
[SPARK-31630][SQL] Fix perf regression by skipping timestamps rebasing after some threshold ### What changes were proposed in this pull request? Skip timestamps rebasing after a global threshold when there is no difference between Julian and Gregorian calendars. This allows to avoid checking hash maps of switch points, and fixes perf regressions in `toJavaTimestamp()` and `fromJavaTimestamp()`. ### Why are the changes needed? The changes fix perf regressions of conversions to/from external type `java.sql.Timestamp`. Before (see the PR's results https://github.com/apache/spark/pull/28440): ``` ================================================================================================ Conversion from/to external types ================================================================================================ OpenJDK 64-Bit Server VM 1.8.0_252-8u252-b09-1~18.04-b09 on Linux 4.15.0-1063-aws Intel(R) Xeon(R) CPU E5-2670 v2 2.50GHz To/from Java's date-time: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative ------------------------------------------------------------------------------------------------------------------------ From java.sql.Timestamp 376 388 10 13.3 75.2 1.1X Collect java.sql.Timestamp 1878 1937 64 2.7 375.6 0.2X ``` After: ``` ================================================================================================ Conversion from/to external types ================================================================================================ OpenJDK 64-Bit Server VM 1.8.0_252-8u252-b09-1~18.04-b09 on Linux 4.15.0-1063-aws Intel(R) Xeon(R) CPU E5-2670 v2 2.50GHz To/from Java's date-time: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative ------------------------------------------------------------------------------------------------------------------------ From java.sql.Timestamp 249 264 24 20.1 49.8 1.7X Collect java.sql.Timestamp 1503 1523 24 3.3 300.5 0.3X ``` Perf improvements in average of: 1. From java.sql.Timestamp is ~ 34% 2. To java.sql.Timestamps is ~16% ### Does this PR introduce _any_ user-facing change? No ### How was this patch tested? By existing test suites `DateTimeUtilsSuite` and `RebaseDateTimeSuite`. Closes #28441 from MaxGekk/opt-rebase-common-threshold. Authored-by: Max Gekk <max.gekk@gmail.com> Signed-off-by: Wenchen Fan <wenchen@databricks.com>
2020-05-05 10:11:53 -04:00
current_date wholestage off 291 293 2 34.3 29.1 1.0X
current_date wholestage on 280 284 3 35.7 28.0 1.0X
[SPARK-29320][TESTS] Compare `sql/core` module in JDK8/11 (Part 1) ### What changes were proposed in this pull request? This PR regenerates the `sql/core` benchmarks in JDK8/11 to compare the result. In general, we compare the ratio instead of the time. However, in this PR, the average time is compared. This PR should be considered as a rough comparison. **A. EXPECTED CASES(JDK11 is faster in general)** - [x] BloomFilterBenchmark (JDK11 is faster except one case) - [x] BuiltInDataSourceWriteBenchmark (JDK11 is faster at CSV/ORC) - [x] CSVBenchmark (JDK11 is faster except five cases) - [x] ColumnarBatchBenchmark (JDK11 is faster at `boolean`/`string` and some cases in `int`/`array`) - [x] DatasetBenchmark (JDK11 is faster with `string`, but is slower for `long` type) - [x] ExternalAppendOnlyUnsafeRowArrayBenchmark (JDK11 is faster except two cases) - [x] ExtractBenchmark (JDK11 is faster except HOUR/MINUTE/SECOND/MILLISECONDS/MICROSECONDS) - [x] HashedRelationMetricsBenchmark (JDK11 is faster) - [x] JSONBenchmark (JDK11 is much faster except eight cases) - [x] JoinBenchmark (JDK11 is faster except five cases) - [x] OrcNestedSchemaPruningBenchmark (JDK11 is faster in nine cases) - [x] PrimitiveArrayBenchmark (JDK11 is faster) - [x] SortBenchmark (JDK11 is faster except `Arrays.sort` case) - [x] UDFBenchmark (N/A, values are too small) - [x] UnsafeArrayDataBenchmark (JDK11 is faster except one case) - [x] WideTableBenchmark (JDK11 is faster except two cases) **B. CASES WE NEED TO INVESTIGATE MORE LATER** - [x] AggregateBenchmark (JDK11 is slower in general) - [x] CompressionSchemeBenchmark (JDK11 is slower in general except `string`) - [x] DataSourceReadBenchmark (JDK11 is slower in general) - [x] DateTimeBenchmark (JDK11 is slightly slower in general except `parsing`) - [x] MakeDateTimeBenchmark (JDK11 is slower except two cases) - [x] MiscBenchmark (JDK11 is slower except ten cases) - [x] OrcV2NestedSchemaPruningBenchmark (JDK11 is slower) - [x] ParquetNestedSchemaPruningBenchmark (JDK11 is slower except six cases) - [x] RangeBenchmark (JDK11 is slower except one case) `FilterPushdownBenchmark/InExpressionBenchmark/WideSchemaBenchmark` will be compared later because it took long timer. ### Why are the changes needed? According to the result, there are some difference between JDK8/JDK11. This will be a baseline for the future improvement and comparison. Also, as a reproducible environment, the following environment is used. - Instance: `r3.xlarge` - OS: `CentOS Linux release 7.5.1804 (Core)` - JDK: - `OpenJDK Runtime Environment (build 1.8.0_222-b10)` - `OpenJDK Runtime Environment 18.9 (build 11.0.4+11-LTS)` ### Does this PR introduce any user-facing change? No. ### How was this patch tested? This is a test-only PR. We need to run benchmark. Closes #26003 from dongjoon-hyun/SPARK-29320. Authored-by: Dongjoon Hyun <dhyun@apple.com> Signed-off-by: Dongjoon Hyun <dhyun@apple.com>
2019-10-03 11:58:25 -04:00
[SPARK-31527][SQL][TESTS][FOLLOWUP] Fix the number of rows in `DateTimeBenchmark` ### What changes were proposed in this pull request? - Changed to the number of rows in benchmark cases from 3 to the actual number `N`. - Regenerated benchmark results in the environment: | Item | Description | | ---- | ----| | Region | us-west-2 (Oregon) | | Instance | r3.xlarge | | AMI | ubuntu/images/hvm-ssd/ubuntu-bionic-18.04-amd64-server-20190722.1 (ami-06f2f779464715dc5) | | Java | OpenJDK 64-Bit Server VM 1.8.0_242 and OpenJDK 64-Bit Server VM 11.0.6+10 | ### Why are the changes needed? The changes are needed to have: - Correct benchmark results - Base line for other perf improvements that can be checked in the same environment. ### Does this PR introduce _any_ user-facing change? No ### How was this patch tested? By running the benchmark and checking its output. Closes #28440 from MaxGekk/SPARK-31527-DateTimeBenchmark-followup. Authored-by: Max Gekk <max.gekk@gmail.com> Signed-off-by: Takeshi Yamamuro <yamamuro@apache.org>
2020-05-03 20:39:50 -04:00
OpenJDK 64-Bit Server VM 11.0.7+10-post-Ubuntu-2ubuntu218.04 on Linux 4.15.0-1063-aws
Intel(R) Xeon(R) CPU E5-2670 v2 @ 2.50GHz
[SPARK-29320][TESTS] Compare `sql/core` module in JDK8/11 (Part 1) ### What changes were proposed in this pull request? This PR regenerates the `sql/core` benchmarks in JDK8/11 to compare the result. In general, we compare the ratio instead of the time. However, in this PR, the average time is compared. This PR should be considered as a rough comparison. **A. EXPECTED CASES(JDK11 is faster in general)** - [x] BloomFilterBenchmark (JDK11 is faster except one case) - [x] BuiltInDataSourceWriteBenchmark (JDK11 is faster at CSV/ORC) - [x] CSVBenchmark (JDK11 is faster except five cases) - [x] ColumnarBatchBenchmark (JDK11 is faster at `boolean`/`string` and some cases in `int`/`array`) - [x] DatasetBenchmark (JDK11 is faster with `string`, but is slower for `long` type) - [x] ExternalAppendOnlyUnsafeRowArrayBenchmark (JDK11 is faster except two cases) - [x] ExtractBenchmark (JDK11 is faster except HOUR/MINUTE/SECOND/MILLISECONDS/MICROSECONDS) - [x] HashedRelationMetricsBenchmark (JDK11 is faster) - [x] JSONBenchmark (JDK11 is much faster except eight cases) - [x] JoinBenchmark (JDK11 is faster except five cases) - [x] OrcNestedSchemaPruningBenchmark (JDK11 is faster in nine cases) - [x] PrimitiveArrayBenchmark (JDK11 is faster) - [x] SortBenchmark (JDK11 is faster except `Arrays.sort` case) - [x] UDFBenchmark (N/A, values are too small) - [x] UnsafeArrayDataBenchmark (JDK11 is faster except one case) - [x] WideTableBenchmark (JDK11 is faster except two cases) **B. CASES WE NEED TO INVESTIGATE MORE LATER** - [x] AggregateBenchmark (JDK11 is slower in general) - [x] CompressionSchemeBenchmark (JDK11 is slower in general except `string`) - [x] DataSourceReadBenchmark (JDK11 is slower in general) - [x] DateTimeBenchmark (JDK11 is slightly slower in general except `parsing`) - [x] MakeDateTimeBenchmark (JDK11 is slower except two cases) - [x] MiscBenchmark (JDK11 is slower except ten cases) - [x] OrcV2NestedSchemaPruningBenchmark (JDK11 is slower) - [x] ParquetNestedSchemaPruningBenchmark (JDK11 is slower except six cases) - [x] RangeBenchmark (JDK11 is slower except one case) `FilterPushdownBenchmark/InExpressionBenchmark/WideSchemaBenchmark` will be compared later because it took long timer. ### Why are the changes needed? According to the result, there are some difference between JDK8/JDK11. This will be a baseline for the future improvement and comparison. Also, as a reproducible environment, the following environment is used. - Instance: `r3.xlarge` - OS: `CentOS Linux release 7.5.1804 (Core)` - JDK: - `OpenJDK Runtime Environment (build 1.8.0_222-b10)` - `OpenJDK Runtime Environment 18.9 (build 11.0.4+11-LTS)` ### Does this PR introduce any user-facing change? No. ### How was this patch tested? This is a test-only PR. We need to run benchmark. Closes #26003 from dongjoon-hyun/SPARK-29320. Authored-by: Dongjoon Hyun <dhyun@apple.com> Signed-off-by: Dongjoon Hyun <dhyun@apple.com>
2019-10-03 11:58:25 -04:00
current_timestamp: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative
------------------------------------------------------------------------------------------------------------------------
[SPARK-31630][SQL] Fix perf regression by skipping timestamps rebasing after some threshold ### What changes were proposed in this pull request? Skip timestamps rebasing after a global threshold when there is no difference between Julian and Gregorian calendars. This allows to avoid checking hash maps of switch points, and fixes perf regressions in `toJavaTimestamp()` and `fromJavaTimestamp()`. ### Why are the changes needed? The changes fix perf regressions of conversions to/from external type `java.sql.Timestamp`. Before (see the PR's results https://github.com/apache/spark/pull/28440): ``` ================================================================================================ Conversion from/to external types ================================================================================================ OpenJDK 64-Bit Server VM 1.8.0_252-8u252-b09-1~18.04-b09 on Linux 4.15.0-1063-aws Intel(R) Xeon(R) CPU E5-2670 v2 2.50GHz To/from Java's date-time: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative ------------------------------------------------------------------------------------------------------------------------ From java.sql.Timestamp 376 388 10 13.3 75.2 1.1X Collect java.sql.Timestamp 1878 1937 64 2.7 375.6 0.2X ``` After: ``` ================================================================================================ Conversion from/to external types ================================================================================================ OpenJDK 64-Bit Server VM 1.8.0_252-8u252-b09-1~18.04-b09 on Linux 4.15.0-1063-aws Intel(R) Xeon(R) CPU E5-2670 v2 2.50GHz To/from Java's date-time: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative ------------------------------------------------------------------------------------------------------------------------ From java.sql.Timestamp 249 264 24 20.1 49.8 1.7X Collect java.sql.Timestamp 1503 1523 24 3.3 300.5 0.3X ``` Perf improvements in average of: 1. From java.sql.Timestamp is ~ 34% 2. To java.sql.Timestamps is ~16% ### Does this PR introduce _any_ user-facing change? No ### How was this patch tested? By existing test suites `DateTimeUtilsSuite` and `RebaseDateTimeSuite`. Closes #28441 from MaxGekk/opt-rebase-common-threshold. Authored-by: Max Gekk <max.gekk@gmail.com> Signed-off-by: Wenchen Fan <wenchen@databricks.com>
2020-05-05 10:11:53 -04:00
current_timestamp wholestage off 311 324 18 32.1 31.1 1.0X
current_timestamp wholestage on 275 364 85 36.3 27.5 1.1X
[SPARK-29320][TESTS] Compare `sql/core` module in JDK8/11 (Part 1) ### What changes were proposed in this pull request? This PR regenerates the `sql/core` benchmarks in JDK8/11 to compare the result. In general, we compare the ratio instead of the time. However, in this PR, the average time is compared. This PR should be considered as a rough comparison. **A. EXPECTED CASES(JDK11 is faster in general)** - [x] BloomFilterBenchmark (JDK11 is faster except one case) - [x] BuiltInDataSourceWriteBenchmark (JDK11 is faster at CSV/ORC) - [x] CSVBenchmark (JDK11 is faster except five cases) - [x] ColumnarBatchBenchmark (JDK11 is faster at `boolean`/`string` and some cases in `int`/`array`) - [x] DatasetBenchmark (JDK11 is faster with `string`, but is slower for `long` type) - [x] ExternalAppendOnlyUnsafeRowArrayBenchmark (JDK11 is faster except two cases) - [x] ExtractBenchmark (JDK11 is faster except HOUR/MINUTE/SECOND/MILLISECONDS/MICROSECONDS) - [x] HashedRelationMetricsBenchmark (JDK11 is faster) - [x] JSONBenchmark (JDK11 is much faster except eight cases) - [x] JoinBenchmark (JDK11 is faster except five cases) - [x] OrcNestedSchemaPruningBenchmark (JDK11 is faster in nine cases) - [x] PrimitiveArrayBenchmark (JDK11 is faster) - [x] SortBenchmark (JDK11 is faster except `Arrays.sort` case) - [x] UDFBenchmark (N/A, values are too small) - [x] UnsafeArrayDataBenchmark (JDK11 is faster except one case) - [x] WideTableBenchmark (JDK11 is faster except two cases) **B. CASES WE NEED TO INVESTIGATE MORE LATER** - [x] AggregateBenchmark (JDK11 is slower in general) - [x] CompressionSchemeBenchmark (JDK11 is slower in general except `string`) - [x] DataSourceReadBenchmark (JDK11 is slower in general) - [x] DateTimeBenchmark (JDK11 is slightly slower in general except `parsing`) - [x] MakeDateTimeBenchmark (JDK11 is slower except two cases) - [x] MiscBenchmark (JDK11 is slower except ten cases) - [x] OrcV2NestedSchemaPruningBenchmark (JDK11 is slower) - [x] ParquetNestedSchemaPruningBenchmark (JDK11 is slower except six cases) - [x] RangeBenchmark (JDK11 is slower except one case) `FilterPushdownBenchmark/InExpressionBenchmark/WideSchemaBenchmark` will be compared later because it took long timer. ### Why are the changes needed? According to the result, there are some difference between JDK8/JDK11. This will be a baseline for the future improvement and comparison. Also, as a reproducible environment, the following environment is used. - Instance: `r3.xlarge` - OS: `CentOS Linux release 7.5.1804 (Core)` - JDK: - `OpenJDK Runtime Environment (build 1.8.0_222-b10)` - `OpenJDK Runtime Environment 18.9 (build 11.0.4+11-LTS)` ### Does this PR introduce any user-facing change? No. ### How was this patch tested? This is a test-only PR. We need to run benchmark. Closes #26003 from dongjoon-hyun/SPARK-29320. Authored-by: Dongjoon Hyun <dhyun@apple.com> Signed-off-by: Dongjoon Hyun <dhyun@apple.com>
2019-10-03 11:58:25 -04:00
================================================================================================
Date arithmetic
================================================================================================
[SPARK-31527][SQL][TESTS][FOLLOWUP] Fix the number of rows in `DateTimeBenchmark` ### What changes were proposed in this pull request? - Changed to the number of rows in benchmark cases from 3 to the actual number `N`. - Regenerated benchmark results in the environment: | Item | Description | | ---- | ----| | Region | us-west-2 (Oregon) | | Instance | r3.xlarge | | AMI | ubuntu/images/hvm-ssd/ubuntu-bionic-18.04-amd64-server-20190722.1 (ami-06f2f779464715dc5) | | Java | OpenJDK 64-Bit Server VM 1.8.0_242 and OpenJDK 64-Bit Server VM 11.0.6+10 | ### Why are the changes needed? The changes are needed to have: - Correct benchmark results - Base line for other perf improvements that can be checked in the same environment. ### Does this PR introduce _any_ user-facing change? No ### How was this patch tested? By running the benchmark and checking its output. Closes #28440 from MaxGekk/SPARK-31527-DateTimeBenchmark-followup. Authored-by: Max Gekk <max.gekk@gmail.com> Signed-off-by: Takeshi Yamamuro <yamamuro@apache.org>
2020-05-03 20:39:50 -04:00
OpenJDK 64-Bit Server VM 11.0.7+10-post-Ubuntu-2ubuntu218.04 on Linux 4.15.0-1063-aws
Intel(R) Xeon(R) CPU E5-2670 v2 @ 2.50GHz
[SPARK-29320][TESTS] Compare `sql/core` module in JDK8/11 (Part 1) ### What changes were proposed in this pull request? This PR regenerates the `sql/core` benchmarks in JDK8/11 to compare the result. In general, we compare the ratio instead of the time. However, in this PR, the average time is compared. This PR should be considered as a rough comparison. **A. EXPECTED CASES(JDK11 is faster in general)** - [x] BloomFilterBenchmark (JDK11 is faster except one case) - [x] BuiltInDataSourceWriteBenchmark (JDK11 is faster at CSV/ORC) - [x] CSVBenchmark (JDK11 is faster except five cases) - [x] ColumnarBatchBenchmark (JDK11 is faster at `boolean`/`string` and some cases in `int`/`array`) - [x] DatasetBenchmark (JDK11 is faster with `string`, but is slower for `long` type) - [x] ExternalAppendOnlyUnsafeRowArrayBenchmark (JDK11 is faster except two cases) - [x] ExtractBenchmark (JDK11 is faster except HOUR/MINUTE/SECOND/MILLISECONDS/MICROSECONDS) - [x] HashedRelationMetricsBenchmark (JDK11 is faster) - [x] JSONBenchmark (JDK11 is much faster except eight cases) - [x] JoinBenchmark (JDK11 is faster except five cases) - [x] OrcNestedSchemaPruningBenchmark (JDK11 is faster in nine cases) - [x] PrimitiveArrayBenchmark (JDK11 is faster) - [x] SortBenchmark (JDK11 is faster except `Arrays.sort` case) - [x] UDFBenchmark (N/A, values are too small) - [x] UnsafeArrayDataBenchmark (JDK11 is faster except one case) - [x] WideTableBenchmark (JDK11 is faster except two cases) **B. CASES WE NEED TO INVESTIGATE MORE LATER** - [x] AggregateBenchmark (JDK11 is slower in general) - [x] CompressionSchemeBenchmark (JDK11 is slower in general except `string`) - [x] DataSourceReadBenchmark (JDK11 is slower in general) - [x] DateTimeBenchmark (JDK11 is slightly slower in general except `parsing`) - [x] MakeDateTimeBenchmark (JDK11 is slower except two cases) - [x] MiscBenchmark (JDK11 is slower except ten cases) - [x] OrcV2NestedSchemaPruningBenchmark (JDK11 is slower) - [x] ParquetNestedSchemaPruningBenchmark (JDK11 is slower except six cases) - [x] RangeBenchmark (JDK11 is slower except one case) `FilterPushdownBenchmark/InExpressionBenchmark/WideSchemaBenchmark` will be compared later because it took long timer. ### Why are the changes needed? According to the result, there are some difference between JDK8/JDK11. This will be a baseline for the future improvement and comparison. Also, as a reproducible environment, the following environment is used. - Instance: `r3.xlarge` - OS: `CentOS Linux release 7.5.1804 (Core)` - JDK: - `OpenJDK Runtime Environment (build 1.8.0_222-b10)` - `OpenJDK Runtime Environment 18.9 (build 11.0.4+11-LTS)` ### Does this PR introduce any user-facing change? No. ### How was this patch tested? This is a test-only PR. We need to run benchmark. Closes #26003 from dongjoon-hyun/SPARK-29320. Authored-by: Dongjoon Hyun <dhyun@apple.com> Signed-off-by: Dongjoon Hyun <dhyun@apple.com>
2019-10-03 11:58:25 -04:00
cast to date: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative
------------------------------------------------------------------------------------------------------------------------
[SPARK-31630][SQL] Fix perf regression by skipping timestamps rebasing after some threshold ### What changes were proposed in this pull request? Skip timestamps rebasing after a global threshold when there is no difference between Julian and Gregorian calendars. This allows to avoid checking hash maps of switch points, and fixes perf regressions in `toJavaTimestamp()` and `fromJavaTimestamp()`. ### Why are the changes needed? The changes fix perf regressions of conversions to/from external type `java.sql.Timestamp`. Before (see the PR's results https://github.com/apache/spark/pull/28440): ``` ================================================================================================ Conversion from/to external types ================================================================================================ OpenJDK 64-Bit Server VM 1.8.0_252-8u252-b09-1~18.04-b09 on Linux 4.15.0-1063-aws Intel(R) Xeon(R) CPU E5-2670 v2 2.50GHz To/from Java's date-time: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative ------------------------------------------------------------------------------------------------------------------------ From java.sql.Timestamp 376 388 10 13.3 75.2 1.1X Collect java.sql.Timestamp 1878 1937 64 2.7 375.6 0.2X ``` After: ``` ================================================================================================ Conversion from/to external types ================================================================================================ OpenJDK 64-Bit Server VM 1.8.0_252-8u252-b09-1~18.04-b09 on Linux 4.15.0-1063-aws Intel(R) Xeon(R) CPU E5-2670 v2 2.50GHz To/from Java's date-time: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative ------------------------------------------------------------------------------------------------------------------------ From java.sql.Timestamp 249 264 24 20.1 49.8 1.7X Collect java.sql.Timestamp 1503 1523 24 3.3 300.5 0.3X ``` Perf improvements in average of: 1. From java.sql.Timestamp is ~ 34% 2. To java.sql.Timestamps is ~16% ### Does this PR introduce _any_ user-facing change? No ### How was this patch tested? By existing test suites `DateTimeUtilsSuite` and `RebaseDateTimeSuite`. Closes #28441 from MaxGekk/opt-rebase-common-threshold. Authored-by: Max Gekk <max.gekk@gmail.com> Signed-off-by: Wenchen Fan <wenchen@databricks.com>
2020-05-05 10:11:53 -04:00
cast to date wholestage off 1077 1079 3 9.3 107.7 1.0X
cast to date wholestage on 1018 1030 14 9.8 101.8 1.1X
[SPARK-29320][TESTS] Compare `sql/core` module in JDK8/11 (Part 1) ### What changes were proposed in this pull request? This PR regenerates the `sql/core` benchmarks in JDK8/11 to compare the result. In general, we compare the ratio instead of the time. However, in this PR, the average time is compared. This PR should be considered as a rough comparison. **A. EXPECTED CASES(JDK11 is faster in general)** - [x] BloomFilterBenchmark (JDK11 is faster except one case) - [x] BuiltInDataSourceWriteBenchmark (JDK11 is faster at CSV/ORC) - [x] CSVBenchmark (JDK11 is faster except five cases) - [x] ColumnarBatchBenchmark (JDK11 is faster at `boolean`/`string` and some cases in `int`/`array`) - [x] DatasetBenchmark (JDK11 is faster with `string`, but is slower for `long` type) - [x] ExternalAppendOnlyUnsafeRowArrayBenchmark (JDK11 is faster except two cases) - [x] ExtractBenchmark (JDK11 is faster except HOUR/MINUTE/SECOND/MILLISECONDS/MICROSECONDS) - [x] HashedRelationMetricsBenchmark (JDK11 is faster) - [x] JSONBenchmark (JDK11 is much faster except eight cases) - [x] JoinBenchmark (JDK11 is faster except five cases) - [x] OrcNestedSchemaPruningBenchmark (JDK11 is faster in nine cases) - [x] PrimitiveArrayBenchmark (JDK11 is faster) - [x] SortBenchmark (JDK11 is faster except `Arrays.sort` case) - [x] UDFBenchmark (N/A, values are too small) - [x] UnsafeArrayDataBenchmark (JDK11 is faster except one case) - [x] WideTableBenchmark (JDK11 is faster except two cases) **B. CASES WE NEED TO INVESTIGATE MORE LATER** - [x] AggregateBenchmark (JDK11 is slower in general) - [x] CompressionSchemeBenchmark (JDK11 is slower in general except `string`) - [x] DataSourceReadBenchmark (JDK11 is slower in general) - [x] DateTimeBenchmark (JDK11 is slightly slower in general except `parsing`) - [x] MakeDateTimeBenchmark (JDK11 is slower except two cases) - [x] MiscBenchmark (JDK11 is slower except ten cases) - [x] OrcV2NestedSchemaPruningBenchmark (JDK11 is slower) - [x] ParquetNestedSchemaPruningBenchmark (JDK11 is slower except six cases) - [x] RangeBenchmark (JDK11 is slower except one case) `FilterPushdownBenchmark/InExpressionBenchmark/WideSchemaBenchmark` will be compared later because it took long timer. ### Why are the changes needed? According to the result, there are some difference between JDK8/JDK11. This will be a baseline for the future improvement and comparison. Also, as a reproducible environment, the following environment is used. - Instance: `r3.xlarge` - OS: `CentOS Linux release 7.5.1804 (Core)` - JDK: - `OpenJDK Runtime Environment (build 1.8.0_222-b10)` - `OpenJDK Runtime Environment 18.9 (build 11.0.4+11-LTS)` ### Does this PR introduce any user-facing change? No. ### How was this patch tested? This is a test-only PR. We need to run benchmark. Closes #26003 from dongjoon-hyun/SPARK-29320. Authored-by: Dongjoon Hyun <dhyun@apple.com> Signed-off-by: Dongjoon Hyun <dhyun@apple.com>
2019-10-03 11:58:25 -04:00
[SPARK-31527][SQL][TESTS][FOLLOWUP] Fix the number of rows in `DateTimeBenchmark` ### What changes were proposed in this pull request? - Changed to the number of rows in benchmark cases from 3 to the actual number `N`. - Regenerated benchmark results in the environment: | Item | Description | | ---- | ----| | Region | us-west-2 (Oregon) | | Instance | r3.xlarge | | AMI | ubuntu/images/hvm-ssd/ubuntu-bionic-18.04-amd64-server-20190722.1 (ami-06f2f779464715dc5) | | Java | OpenJDK 64-Bit Server VM 1.8.0_242 and OpenJDK 64-Bit Server VM 11.0.6+10 | ### Why are the changes needed? The changes are needed to have: - Correct benchmark results - Base line for other perf improvements that can be checked in the same environment. ### Does this PR introduce _any_ user-facing change? No ### How was this patch tested? By running the benchmark and checking its output. Closes #28440 from MaxGekk/SPARK-31527-DateTimeBenchmark-followup. Authored-by: Max Gekk <max.gekk@gmail.com> Signed-off-by: Takeshi Yamamuro <yamamuro@apache.org>
2020-05-03 20:39:50 -04:00
OpenJDK 64-Bit Server VM 11.0.7+10-post-Ubuntu-2ubuntu218.04 on Linux 4.15.0-1063-aws
Intel(R) Xeon(R) CPU E5-2670 v2 @ 2.50GHz
[SPARK-29320][TESTS] Compare `sql/core` module in JDK8/11 (Part 1) ### What changes were proposed in this pull request? This PR regenerates the `sql/core` benchmarks in JDK8/11 to compare the result. In general, we compare the ratio instead of the time. However, in this PR, the average time is compared. This PR should be considered as a rough comparison. **A. EXPECTED CASES(JDK11 is faster in general)** - [x] BloomFilterBenchmark (JDK11 is faster except one case) - [x] BuiltInDataSourceWriteBenchmark (JDK11 is faster at CSV/ORC) - [x] CSVBenchmark (JDK11 is faster except five cases) - [x] ColumnarBatchBenchmark (JDK11 is faster at `boolean`/`string` and some cases in `int`/`array`) - [x] DatasetBenchmark (JDK11 is faster with `string`, but is slower for `long` type) - [x] ExternalAppendOnlyUnsafeRowArrayBenchmark (JDK11 is faster except two cases) - [x] ExtractBenchmark (JDK11 is faster except HOUR/MINUTE/SECOND/MILLISECONDS/MICROSECONDS) - [x] HashedRelationMetricsBenchmark (JDK11 is faster) - [x] JSONBenchmark (JDK11 is much faster except eight cases) - [x] JoinBenchmark (JDK11 is faster except five cases) - [x] OrcNestedSchemaPruningBenchmark (JDK11 is faster in nine cases) - [x] PrimitiveArrayBenchmark (JDK11 is faster) - [x] SortBenchmark (JDK11 is faster except `Arrays.sort` case) - [x] UDFBenchmark (N/A, values are too small) - [x] UnsafeArrayDataBenchmark (JDK11 is faster except one case) - [x] WideTableBenchmark (JDK11 is faster except two cases) **B. CASES WE NEED TO INVESTIGATE MORE LATER** - [x] AggregateBenchmark (JDK11 is slower in general) - [x] CompressionSchemeBenchmark (JDK11 is slower in general except `string`) - [x] DataSourceReadBenchmark (JDK11 is slower in general) - [x] DateTimeBenchmark (JDK11 is slightly slower in general except `parsing`) - [x] MakeDateTimeBenchmark (JDK11 is slower except two cases) - [x] MiscBenchmark (JDK11 is slower except ten cases) - [x] OrcV2NestedSchemaPruningBenchmark (JDK11 is slower) - [x] ParquetNestedSchemaPruningBenchmark (JDK11 is slower except six cases) - [x] RangeBenchmark (JDK11 is slower except one case) `FilterPushdownBenchmark/InExpressionBenchmark/WideSchemaBenchmark` will be compared later because it took long timer. ### Why are the changes needed? According to the result, there are some difference between JDK8/JDK11. This will be a baseline for the future improvement and comparison. Also, as a reproducible environment, the following environment is used. - Instance: `r3.xlarge` - OS: `CentOS Linux release 7.5.1804 (Core)` - JDK: - `OpenJDK Runtime Environment (build 1.8.0_222-b10)` - `OpenJDK Runtime Environment 18.9 (build 11.0.4+11-LTS)` ### Does this PR introduce any user-facing change? No. ### How was this patch tested? This is a test-only PR. We need to run benchmark. Closes #26003 from dongjoon-hyun/SPARK-29320. Authored-by: Dongjoon Hyun <dhyun@apple.com> Signed-off-by: Dongjoon Hyun <dhyun@apple.com>
2019-10-03 11:58:25 -04:00
last_day: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative
------------------------------------------------------------------------------------------------------------------------
[SPARK-31630][SQL] Fix perf regression by skipping timestamps rebasing after some threshold ### What changes were proposed in this pull request? Skip timestamps rebasing after a global threshold when there is no difference between Julian and Gregorian calendars. This allows to avoid checking hash maps of switch points, and fixes perf regressions in `toJavaTimestamp()` and `fromJavaTimestamp()`. ### Why are the changes needed? The changes fix perf regressions of conversions to/from external type `java.sql.Timestamp`. Before (see the PR's results https://github.com/apache/spark/pull/28440): ``` ================================================================================================ Conversion from/to external types ================================================================================================ OpenJDK 64-Bit Server VM 1.8.0_252-8u252-b09-1~18.04-b09 on Linux 4.15.0-1063-aws Intel(R) Xeon(R) CPU E5-2670 v2 2.50GHz To/from Java's date-time: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative ------------------------------------------------------------------------------------------------------------------------ From java.sql.Timestamp 376 388 10 13.3 75.2 1.1X Collect java.sql.Timestamp 1878 1937 64 2.7 375.6 0.2X ``` After: ``` ================================================================================================ Conversion from/to external types ================================================================================================ OpenJDK 64-Bit Server VM 1.8.0_252-8u252-b09-1~18.04-b09 on Linux 4.15.0-1063-aws Intel(R) Xeon(R) CPU E5-2670 v2 2.50GHz To/from Java's date-time: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative ------------------------------------------------------------------------------------------------------------------------ From java.sql.Timestamp 249 264 24 20.1 49.8 1.7X Collect java.sql.Timestamp 1503 1523 24 3.3 300.5 0.3X ``` Perf improvements in average of: 1. From java.sql.Timestamp is ~ 34% 2. To java.sql.Timestamps is ~16% ### Does this PR introduce _any_ user-facing change? No ### How was this patch tested? By existing test suites `DateTimeUtilsSuite` and `RebaseDateTimeSuite`. Closes #28441 from MaxGekk/opt-rebase-common-threshold. Authored-by: Max Gekk <max.gekk@gmail.com> Signed-off-by: Wenchen Fan <wenchen@databricks.com>
2020-05-05 10:11:53 -04:00
last_day wholestage off 1257 1260 4 8.0 125.7 1.0X
last_day wholestage on 1218 1227 14 8.2 121.8 1.0X
[SPARK-29320][TESTS] Compare `sql/core` module in JDK8/11 (Part 1) ### What changes were proposed in this pull request? This PR regenerates the `sql/core` benchmarks in JDK8/11 to compare the result. In general, we compare the ratio instead of the time. However, in this PR, the average time is compared. This PR should be considered as a rough comparison. **A. EXPECTED CASES(JDK11 is faster in general)** - [x] BloomFilterBenchmark (JDK11 is faster except one case) - [x] BuiltInDataSourceWriteBenchmark (JDK11 is faster at CSV/ORC) - [x] CSVBenchmark (JDK11 is faster except five cases) - [x] ColumnarBatchBenchmark (JDK11 is faster at `boolean`/`string` and some cases in `int`/`array`) - [x] DatasetBenchmark (JDK11 is faster with `string`, but is slower for `long` type) - [x] ExternalAppendOnlyUnsafeRowArrayBenchmark (JDK11 is faster except two cases) - [x] ExtractBenchmark (JDK11 is faster except HOUR/MINUTE/SECOND/MILLISECONDS/MICROSECONDS) - [x] HashedRelationMetricsBenchmark (JDK11 is faster) - [x] JSONBenchmark (JDK11 is much faster except eight cases) - [x] JoinBenchmark (JDK11 is faster except five cases) - [x] OrcNestedSchemaPruningBenchmark (JDK11 is faster in nine cases) - [x] PrimitiveArrayBenchmark (JDK11 is faster) - [x] SortBenchmark (JDK11 is faster except `Arrays.sort` case) - [x] UDFBenchmark (N/A, values are too small) - [x] UnsafeArrayDataBenchmark (JDK11 is faster except one case) - [x] WideTableBenchmark (JDK11 is faster except two cases) **B. CASES WE NEED TO INVESTIGATE MORE LATER** - [x] AggregateBenchmark (JDK11 is slower in general) - [x] CompressionSchemeBenchmark (JDK11 is slower in general except `string`) - [x] DataSourceReadBenchmark (JDK11 is slower in general) - [x] DateTimeBenchmark (JDK11 is slightly slower in general except `parsing`) - [x] MakeDateTimeBenchmark (JDK11 is slower except two cases) - [x] MiscBenchmark (JDK11 is slower except ten cases) - [x] OrcV2NestedSchemaPruningBenchmark (JDK11 is slower) - [x] ParquetNestedSchemaPruningBenchmark (JDK11 is slower except six cases) - [x] RangeBenchmark (JDK11 is slower except one case) `FilterPushdownBenchmark/InExpressionBenchmark/WideSchemaBenchmark` will be compared later because it took long timer. ### Why are the changes needed? According to the result, there are some difference between JDK8/JDK11. This will be a baseline for the future improvement and comparison. Also, as a reproducible environment, the following environment is used. - Instance: `r3.xlarge` - OS: `CentOS Linux release 7.5.1804 (Core)` - JDK: - `OpenJDK Runtime Environment (build 1.8.0_222-b10)` - `OpenJDK Runtime Environment 18.9 (build 11.0.4+11-LTS)` ### Does this PR introduce any user-facing change? No. ### How was this patch tested? This is a test-only PR. We need to run benchmark. Closes #26003 from dongjoon-hyun/SPARK-29320. Authored-by: Dongjoon Hyun <dhyun@apple.com> Signed-off-by: Dongjoon Hyun <dhyun@apple.com>
2019-10-03 11:58:25 -04:00
[SPARK-31527][SQL][TESTS][FOLLOWUP] Fix the number of rows in `DateTimeBenchmark` ### What changes were proposed in this pull request? - Changed to the number of rows in benchmark cases from 3 to the actual number `N`. - Regenerated benchmark results in the environment: | Item | Description | | ---- | ----| | Region | us-west-2 (Oregon) | | Instance | r3.xlarge | | AMI | ubuntu/images/hvm-ssd/ubuntu-bionic-18.04-amd64-server-20190722.1 (ami-06f2f779464715dc5) | | Java | OpenJDK 64-Bit Server VM 1.8.0_242 and OpenJDK 64-Bit Server VM 11.0.6+10 | ### Why are the changes needed? The changes are needed to have: - Correct benchmark results - Base line for other perf improvements that can be checked in the same environment. ### Does this PR introduce _any_ user-facing change? No ### How was this patch tested? By running the benchmark and checking its output. Closes #28440 from MaxGekk/SPARK-31527-DateTimeBenchmark-followup. Authored-by: Max Gekk <max.gekk@gmail.com> Signed-off-by: Takeshi Yamamuro <yamamuro@apache.org>
2020-05-03 20:39:50 -04:00
OpenJDK 64-Bit Server VM 11.0.7+10-post-Ubuntu-2ubuntu218.04 on Linux 4.15.0-1063-aws
Intel(R) Xeon(R) CPU E5-2670 v2 @ 2.50GHz
[SPARK-29320][TESTS] Compare `sql/core` module in JDK8/11 (Part 1) ### What changes were proposed in this pull request? This PR regenerates the `sql/core` benchmarks in JDK8/11 to compare the result. In general, we compare the ratio instead of the time. However, in this PR, the average time is compared. This PR should be considered as a rough comparison. **A. EXPECTED CASES(JDK11 is faster in general)** - [x] BloomFilterBenchmark (JDK11 is faster except one case) - [x] BuiltInDataSourceWriteBenchmark (JDK11 is faster at CSV/ORC) - [x] CSVBenchmark (JDK11 is faster except five cases) - [x] ColumnarBatchBenchmark (JDK11 is faster at `boolean`/`string` and some cases in `int`/`array`) - [x] DatasetBenchmark (JDK11 is faster with `string`, but is slower for `long` type) - [x] ExternalAppendOnlyUnsafeRowArrayBenchmark (JDK11 is faster except two cases) - [x] ExtractBenchmark (JDK11 is faster except HOUR/MINUTE/SECOND/MILLISECONDS/MICROSECONDS) - [x] HashedRelationMetricsBenchmark (JDK11 is faster) - [x] JSONBenchmark (JDK11 is much faster except eight cases) - [x] JoinBenchmark (JDK11 is faster except five cases) - [x] OrcNestedSchemaPruningBenchmark (JDK11 is faster in nine cases) - [x] PrimitiveArrayBenchmark (JDK11 is faster) - [x] SortBenchmark (JDK11 is faster except `Arrays.sort` case) - [x] UDFBenchmark (N/A, values are too small) - [x] UnsafeArrayDataBenchmark (JDK11 is faster except one case) - [x] WideTableBenchmark (JDK11 is faster except two cases) **B. CASES WE NEED TO INVESTIGATE MORE LATER** - [x] AggregateBenchmark (JDK11 is slower in general) - [x] CompressionSchemeBenchmark (JDK11 is slower in general except `string`) - [x] DataSourceReadBenchmark (JDK11 is slower in general) - [x] DateTimeBenchmark (JDK11 is slightly slower in general except `parsing`) - [x] MakeDateTimeBenchmark (JDK11 is slower except two cases) - [x] MiscBenchmark (JDK11 is slower except ten cases) - [x] OrcV2NestedSchemaPruningBenchmark (JDK11 is slower) - [x] ParquetNestedSchemaPruningBenchmark (JDK11 is slower except six cases) - [x] RangeBenchmark (JDK11 is slower except one case) `FilterPushdownBenchmark/InExpressionBenchmark/WideSchemaBenchmark` will be compared later because it took long timer. ### Why are the changes needed? According to the result, there are some difference between JDK8/JDK11. This will be a baseline for the future improvement and comparison. Also, as a reproducible environment, the following environment is used. - Instance: `r3.xlarge` - OS: `CentOS Linux release 7.5.1804 (Core)` - JDK: - `OpenJDK Runtime Environment (build 1.8.0_222-b10)` - `OpenJDK Runtime Environment 18.9 (build 11.0.4+11-LTS)` ### Does this PR introduce any user-facing change? No. ### How was this patch tested? This is a test-only PR. We need to run benchmark. Closes #26003 from dongjoon-hyun/SPARK-29320. Authored-by: Dongjoon Hyun <dhyun@apple.com> Signed-off-by: Dongjoon Hyun <dhyun@apple.com>
2019-10-03 11:58:25 -04:00
next_day: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative
------------------------------------------------------------------------------------------------------------------------
[SPARK-31630][SQL] Fix perf regression by skipping timestamps rebasing after some threshold ### What changes were proposed in this pull request? Skip timestamps rebasing after a global threshold when there is no difference between Julian and Gregorian calendars. This allows to avoid checking hash maps of switch points, and fixes perf regressions in `toJavaTimestamp()` and `fromJavaTimestamp()`. ### Why are the changes needed? The changes fix perf regressions of conversions to/from external type `java.sql.Timestamp`. Before (see the PR's results https://github.com/apache/spark/pull/28440): ``` ================================================================================================ Conversion from/to external types ================================================================================================ OpenJDK 64-Bit Server VM 1.8.0_252-8u252-b09-1~18.04-b09 on Linux 4.15.0-1063-aws Intel(R) Xeon(R) CPU E5-2670 v2 2.50GHz To/from Java's date-time: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative ------------------------------------------------------------------------------------------------------------------------ From java.sql.Timestamp 376 388 10 13.3 75.2 1.1X Collect java.sql.Timestamp 1878 1937 64 2.7 375.6 0.2X ``` After: ``` ================================================================================================ Conversion from/to external types ================================================================================================ OpenJDK 64-Bit Server VM 1.8.0_252-8u252-b09-1~18.04-b09 on Linux 4.15.0-1063-aws Intel(R) Xeon(R) CPU E5-2670 v2 2.50GHz To/from Java's date-time: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative ------------------------------------------------------------------------------------------------------------------------ From java.sql.Timestamp 249 264 24 20.1 49.8 1.7X Collect java.sql.Timestamp 1503 1523 24 3.3 300.5 0.3X ``` Perf improvements in average of: 1. From java.sql.Timestamp is ~ 34% 2. To java.sql.Timestamps is ~16% ### Does this PR introduce _any_ user-facing change? No ### How was this patch tested? By existing test suites `DateTimeUtilsSuite` and `RebaseDateTimeSuite`. Closes #28441 from MaxGekk/opt-rebase-common-threshold. Authored-by: Max Gekk <max.gekk@gmail.com> Signed-off-by: Wenchen Fan <wenchen@databricks.com>
2020-05-05 10:11:53 -04:00
next_day wholestage off 1140 1141 1 8.8 114.0 1.0X
next_day wholestage on 1067 1076 11 9.4 106.7 1.1X
[SPARK-29320][TESTS] Compare `sql/core` module in JDK8/11 (Part 1) ### What changes were proposed in this pull request? This PR regenerates the `sql/core` benchmarks in JDK8/11 to compare the result. In general, we compare the ratio instead of the time. However, in this PR, the average time is compared. This PR should be considered as a rough comparison. **A. EXPECTED CASES(JDK11 is faster in general)** - [x] BloomFilterBenchmark (JDK11 is faster except one case) - [x] BuiltInDataSourceWriteBenchmark (JDK11 is faster at CSV/ORC) - [x] CSVBenchmark (JDK11 is faster except five cases) - [x] ColumnarBatchBenchmark (JDK11 is faster at `boolean`/`string` and some cases in `int`/`array`) - [x] DatasetBenchmark (JDK11 is faster with `string`, but is slower for `long` type) - [x] ExternalAppendOnlyUnsafeRowArrayBenchmark (JDK11 is faster except two cases) - [x] ExtractBenchmark (JDK11 is faster except HOUR/MINUTE/SECOND/MILLISECONDS/MICROSECONDS) - [x] HashedRelationMetricsBenchmark (JDK11 is faster) - [x] JSONBenchmark (JDK11 is much faster except eight cases) - [x] JoinBenchmark (JDK11 is faster except five cases) - [x] OrcNestedSchemaPruningBenchmark (JDK11 is faster in nine cases) - [x] PrimitiveArrayBenchmark (JDK11 is faster) - [x] SortBenchmark (JDK11 is faster except `Arrays.sort` case) - [x] UDFBenchmark (N/A, values are too small) - [x] UnsafeArrayDataBenchmark (JDK11 is faster except one case) - [x] WideTableBenchmark (JDK11 is faster except two cases) **B. CASES WE NEED TO INVESTIGATE MORE LATER** - [x] AggregateBenchmark (JDK11 is slower in general) - [x] CompressionSchemeBenchmark (JDK11 is slower in general except `string`) - [x] DataSourceReadBenchmark (JDK11 is slower in general) - [x] DateTimeBenchmark (JDK11 is slightly slower in general except `parsing`) - [x] MakeDateTimeBenchmark (JDK11 is slower except two cases) - [x] MiscBenchmark (JDK11 is slower except ten cases) - [x] OrcV2NestedSchemaPruningBenchmark (JDK11 is slower) - [x] ParquetNestedSchemaPruningBenchmark (JDK11 is slower except six cases) - [x] RangeBenchmark (JDK11 is slower except one case) `FilterPushdownBenchmark/InExpressionBenchmark/WideSchemaBenchmark` will be compared later because it took long timer. ### Why are the changes needed? According to the result, there are some difference between JDK8/JDK11. This will be a baseline for the future improvement and comparison. Also, as a reproducible environment, the following environment is used. - Instance: `r3.xlarge` - OS: `CentOS Linux release 7.5.1804 (Core)` - JDK: - `OpenJDK Runtime Environment (build 1.8.0_222-b10)` - `OpenJDK Runtime Environment 18.9 (build 11.0.4+11-LTS)` ### Does this PR introduce any user-facing change? No. ### How was this patch tested? This is a test-only PR. We need to run benchmark. Closes #26003 from dongjoon-hyun/SPARK-29320. Authored-by: Dongjoon Hyun <dhyun@apple.com> Signed-off-by: Dongjoon Hyun <dhyun@apple.com>
2019-10-03 11:58:25 -04:00
[SPARK-31527][SQL][TESTS][FOLLOWUP] Fix the number of rows in `DateTimeBenchmark` ### What changes were proposed in this pull request? - Changed to the number of rows in benchmark cases from 3 to the actual number `N`. - Regenerated benchmark results in the environment: | Item | Description | | ---- | ----| | Region | us-west-2 (Oregon) | | Instance | r3.xlarge | | AMI | ubuntu/images/hvm-ssd/ubuntu-bionic-18.04-amd64-server-20190722.1 (ami-06f2f779464715dc5) | | Java | OpenJDK 64-Bit Server VM 1.8.0_242 and OpenJDK 64-Bit Server VM 11.0.6+10 | ### Why are the changes needed? The changes are needed to have: - Correct benchmark results - Base line for other perf improvements that can be checked in the same environment. ### Does this PR introduce _any_ user-facing change? No ### How was this patch tested? By running the benchmark and checking its output. Closes #28440 from MaxGekk/SPARK-31527-DateTimeBenchmark-followup. Authored-by: Max Gekk <max.gekk@gmail.com> Signed-off-by: Takeshi Yamamuro <yamamuro@apache.org>
2020-05-03 20:39:50 -04:00
OpenJDK 64-Bit Server VM 11.0.7+10-post-Ubuntu-2ubuntu218.04 on Linux 4.15.0-1063-aws
Intel(R) Xeon(R) CPU E5-2670 v2 @ 2.50GHz
[SPARK-29320][TESTS] Compare `sql/core` module in JDK8/11 (Part 1) ### What changes were proposed in this pull request? This PR regenerates the `sql/core` benchmarks in JDK8/11 to compare the result. In general, we compare the ratio instead of the time. However, in this PR, the average time is compared. This PR should be considered as a rough comparison. **A. EXPECTED CASES(JDK11 is faster in general)** - [x] BloomFilterBenchmark (JDK11 is faster except one case) - [x] BuiltInDataSourceWriteBenchmark (JDK11 is faster at CSV/ORC) - [x] CSVBenchmark (JDK11 is faster except five cases) - [x] ColumnarBatchBenchmark (JDK11 is faster at `boolean`/`string` and some cases in `int`/`array`) - [x] DatasetBenchmark (JDK11 is faster with `string`, but is slower for `long` type) - [x] ExternalAppendOnlyUnsafeRowArrayBenchmark (JDK11 is faster except two cases) - [x] ExtractBenchmark (JDK11 is faster except HOUR/MINUTE/SECOND/MILLISECONDS/MICROSECONDS) - [x] HashedRelationMetricsBenchmark (JDK11 is faster) - [x] JSONBenchmark (JDK11 is much faster except eight cases) - [x] JoinBenchmark (JDK11 is faster except five cases) - [x] OrcNestedSchemaPruningBenchmark (JDK11 is faster in nine cases) - [x] PrimitiveArrayBenchmark (JDK11 is faster) - [x] SortBenchmark (JDK11 is faster except `Arrays.sort` case) - [x] UDFBenchmark (N/A, values are too small) - [x] UnsafeArrayDataBenchmark (JDK11 is faster except one case) - [x] WideTableBenchmark (JDK11 is faster except two cases) **B. CASES WE NEED TO INVESTIGATE MORE LATER** - [x] AggregateBenchmark (JDK11 is slower in general) - [x] CompressionSchemeBenchmark (JDK11 is slower in general except `string`) - [x] DataSourceReadBenchmark (JDK11 is slower in general) - [x] DateTimeBenchmark (JDK11 is slightly slower in general except `parsing`) - [x] MakeDateTimeBenchmark (JDK11 is slower except two cases) - [x] MiscBenchmark (JDK11 is slower except ten cases) - [x] OrcV2NestedSchemaPruningBenchmark (JDK11 is slower) - [x] ParquetNestedSchemaPruningBenchmark (JDK11 is slower except six cases) - [x] RangeBenchmark (JDK11 is slower except one case) `FilterPushdownBenchmark/InExpressionBenchmark/WideSchemaBenchmark` will be compared later because it took long timer. ### Why are the changes needed? According to the result, there are some difference between JDK8/JDK11. This will be a baseline for the future improvement and comparison. Also, as a reproducible environment, the following environment is used. - Instance: `r3.xlarge` - OS: `CentOS Linux release 7.5.1804 (Core)` - JDK: - `OpenJDK Runtime Environment (build 1.8.0_222-b10)` - `OpenJDK Runtime Environment 18.9 (build 11.0.4+11-LTS)` ### Does this PR introduce any user-facing change? No. ### How was this patch tested? This is a test-only PR. We need to run benchmark. Closes #26003 from dongjoon-hyun/SPARK-29320. Authored-by: Dongjoon Hyun <dhyun@apple.com> Signed-off-by: Dongjoon Hyun <dhyun@apple.com>
2019-10-03 11:58:25 -04:00
date_add: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative
------------------------------------------------------------------------------------------------------------------------
[SPARK-31630][SQL] Fix perf regression by skipping timestamps rebasing after some threshold ### What changes were proposed in this pull request? Skip timestamps rebasing after a global threshold when there is no difference between Julian and Gregorian calendars. This allows to avoid checking hash maps of switch points, and fixes perf regressions in `toJavaTimestamp()` and `fromJavaTimestamp()`. ### Why are the changes needed? The changes fix perf regressions of conversions to/from external type `java.sql.Timestamp`. Before (see the PR's results https://github.com/apache/spark/pull/28440): ``` ================================================================================================ Conversion from/to external types ================================================================================================ OpenJDK 64-Bit Server VM 1.8.0_252-8u252-b09-1~18.04-b09 on Linux 4.15.0-1063-aws Intel(R) Xeon(R) CPU E5-2670 v2 2.50GHz To/from Java's date-time: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative ------------------------------------------------------------------------------------------------------------------------ From java.sql.Timestamp 376 388 10 13.3 75.2 1.1X Collect java.sql.Timestamp 1878 1937 64 2.7 375.6 0.2X ``` After: ``` ================================================================================================ Conversion from/to external types ================================================================================================ OpenJDK 64-Bit Server VM 1.8.0_252-8u252-b09-1~18.04-b09 on Linux 4.15.0-1063-aws Intel(R) Xeon(R) CPU E5-2670 v2 2.50GHz To/from Java's date-time: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative ------------------------------------------------------------------------------------------------------------------------ From java.sql.Timestamp 249 264 24 20.1 49.8 1.7X Collect java.sql.Timestamp 1503 1523 24 3.3 300.5 0.3X ``` Perf improvements in average of: 1. From java.sql.Timestamp is ~ 34% 2. To java.sql.Timestamps is ~16% ### Does this PR introduce _any_ user-facing change? No ### How was this patch tested? By existing test suites `DateTimeUtilsSuite` and `RebaseDateTimeSuite`. Closes #28441 from MaxGekk/opt-rebase-common-threshold. Authored-by: Max Gekk <max.gekk@gmail.com> Signed-off-by: Wenchen Fan <wenchen@databricks.com>
2020-05-05 10:11:53 -04:00
date_add wholestage off 1062 1064 3 9.4 106.2 1.0X
date_add wholestage on 1046 1055 11 9.6 104.6 1.0X
[SPARK-29320][TESTS] Compare `sql/core` module in JDK8/11 (Part 1) ### What changes were proposed in this pull request? This PR regenerates the `sql/core` benchmarks in JDK8/11 to compare the result. In general, we compare the ratio instead of the time. However, in this PR, the average time is compared. This PR should be considered as a rough comparison. **A. EXPECTED CASES(JDK11 is faster in general)** - [x] BloomFilterBenchmark (JDK11 is faster except one case) - [x] BuiltInDataSourceWriteBenchmark (JDK11 is faster at CSV/ORC) - [x] CSVBenchmark (JDK11 is faster except five cases) - [x] ColumnarBatchBenchmark (JDK11 is faster at `boolean`/`string` and some cases in `int`/`array`) - [x] DatasetBenchmark (JDK11 is faster with `string`, but is slower for `long` type) - [x] ExternalAppendOnlyUnsafeRowArrayBenchmark (JDK11 is faster except two cases) - [x] ExtractBenchmark (JDK11 is faster except HOUR/MINUTE/SECOND/MILLISECONDS/MICROSECONDS) - [x] HashedRelationMetricsBenchmark (JDK11 is faster) - [x] JSONBenchmark (JDK11 is much faster except eight cases) - [x] JoinBenchmark (JDK11 is faster except five cases) - [x] OrcNestedSchemaPruningBenchmark (JDK11 is faster in nine cases) - [x] PrimitiveArrayBenchmark (JDK11 is faster) - [x] SortBenchmark (JDK11 is faster except `Arrays.sort` case) - [x] UDFBenchmark (N/A, values are too small) - [x] UnsafeArrayDataBenchmark (JDK11 is faster except one case) - [x] WideTableBenchmark (JDK11 is faster except two cases) **B. CASES WE NEED TO INVESTIGATE MORE LATER** - [x] AggregateBenchmark (JDK11 is slower in general) - [x] CompressionSchemeBenchmark (JDK11 is slower in general except `string`) - [x] DataSourceReadBenchmark (JDK11 is slower in general) - [x] DateTimeBenchmark (JDK11 is slightly slower in general except `parsing`) - [x] MakeDateTimeBenchmark (JDK11 is slower except two cases) - [x] MiscBenchmark (JDK11 is slower except ten cases) - [x] OrcV2NestedSchemaPruningBenchmark (JDK11 is slower) - [x] ParquetNestedSchemaPruningBenchmark (JDK11 is slower except six cases) - [x] RangeBenchmark (JDK11 is slower except one case) `FilterPushdownBenchmark/InExpressionBenchmark/WideSchemaBenchmark` will be compared later because it took long timer. ### Why are the changes needed? According to the result, there are some difference between JDK8/JDK11. This will be a baseline for the future improvement and comparison. Also, as a reproducible environment, the following environment is used. - Instance: `r3.xlarge` - OS: `CentOS Linux release 7.5.1804 (Core)` - JDK: - `OpenJDK Runtime Environment (build 1.8.0_222-b10)` - `OpenJDK Runtime Environment 18.9 (build 11.0.4+11-LTS)` ### Does this PR introduce any user-facing change? No. ### How was this patch tested? This is a test-only PR. We need to run benchmark. Closes #26003 from dongjoon-hyun/SPARK-29320. Authored-by: Dongjoon Hyun <dhyun@apple.com> Signed-off-by: Dongjoon Hyun <dhyun@apple.com>
2019-10-03 11:58:25 -04:00
[SPARK-31527][SQL][TESTS][FOLLOWUP] Fix the number of rows in `DateTimeBenchmark` ### What changes were proposed in this pull request? - Changed to the number of rows in benchmark cases from 3 to the actual number `N`. - Regenerated benchmark results in the environment: | Item | Description | | ---- | ----| | Region | us-west-2 (Oregon) | | Instance | r3.xlarge | | AMI | ubuntu/images/hvm-ssd/ubuntu-bionic-18.04-amd64-server-20190722.1 (ami-06f2f779464715dc5) | | Java | OpenJDK 64-Bit Server VM 1.8.0_242 and OpenJDK 64-Bit Server VM 11.0.6+10 | ### Why are the changes needed? The changes are needed to have: - Correct benchmark results - Base line for other perf improvements that can be checked in the same environment. ### Does this PR introduce _any_ user-facing change? No ### How was this patch tested? By running the benchmark and checking its output. Closes #28440 from MaxGekk/SPARK-31527-DateTimeBenchmark-followup. Authored-by: Max Gekk <max.gekk@gmail.com> Signed-off-by: Takeshi Yamamuro <yamamuro@apache.org>
2020-05-03 20:39:50 -04:00
OpenJDK 64-Bit Server VM 11.0.7+10-post-Ubuntu-2ubuntu218.04 on Linux 4.15.0-1063-aws
Intel(R) Xeon(R) CPU E5-2670 v2 @ 2.50GHz
[SPARK-29320][TESTS] Compare `sql/core` module in JDK8/11 (Part 1) ### What changes were proposed in this pull request? This PR regenerates the `sql/core` benchmarks in JDK8/11 to compare the result. In general, we compare the ratio instead of the time. However, in this PR, the average time is compared. This PR should be considered as a rough comparison. **A. EXPECTED CASES(JDK11 is faster in general)** - [x] BloomFilterBenchmark (JDK11 is faster except one case) - [x] BuiltInDataSourceWriteBenchmark (JDK11 is faster at CSV/ORC) - [x] CSVBenchmark (JDK11 is faster except five cases) - [x] ColumnarBatchBenchmark (JDK11 is faster at `boolean`/`string` and some cases in `int`/`array`) - [x] DatasetBenchmark (JDK11 is faster with `string`, but is slower for `long` type) - [x] ExternalAppendOnlyUnsafeRowArrayBenchmark (JDK11 is faster except two cases) - [x] ExtractBenchmark (JDK11 is faster except HOUR/MINUTE/SECOND/MILLISECONDS/MICROSECONDS) - [x] HashedRelationMetricsBenchmark (JDK11 is faster) - [x] JSONBenchmark (JDK11 is much faster except eight cases) - [x] JoinBenchmark (JDK11 is faster except five cases) - [x] OrcNestedSchemaPruningBenchmark (JDK11 is faster in nine cases) - [x] PrimitiveArrayBenchmark (JDK11 is faster) - [x] SortBenchmark (JDK11 is faster except `Arrays.sort` case) - [x] UDFBenchmark (N/A, values are too small) - [x] UnsafeArrayDataBenchmark (JDK11 is faster except one case) - [x] WideTableBenchmark (JDK11 is faster except two cases) **B. CASES WE NEED TO INVESTIGATE MORE LATER** - [x] AggregateBenchmark (JDK11 is slower in general) - [x] CompressionSchemeBenchmark (JDK11 is slower in general except `string`) - [x] DataSourceReadBenchmark (JDK11 is slower in general) - [x] DateTimeBenchmark (JDK11 is slightly slower in general except `parsing`) - [x] MakeDateTimeBenchmark (JDK11 is slower except two cases) - [x] MiscBenchmark (JDK11 is slower except ten cases) - [x] OrcV2NestedSchemaPruningBenchmark (JDK11 is slower) - [x] ParquetNestedSchemaPruningBenchmark (JDK11 is slower except six cases) - [x] RangeBenchmark (JDK11 is slower except one case) `FilterPushdownBenchmark/InExpressionBenchmark/WideSchemaBenchmark` will be compared later because it took long timer. ### Why are the changes needed? According to the result, there are some difference between JDK8/JDK11. This will be a baseline for the future improvement and comparison. Also, as a reproducible environment, the following environment is used. - Instance: `r3.xlarge` - OS: `CentOS Linux release 7.5.1804 (Core)` - JDK: - `OpenJDK Runtime Environment (build 1.8.0_222-b10)` - `OpenJDK Runtime Environment 18.9 (build 11.0.4+11-LTS)` ### Does this PR introduce any user-facing change? No. ### How was this patch tested? This is a test-only PR. We need to run benchmark. Closes #26003 from dongjoon-hyun/SPARK-29320. Authored-by: Dongjoon Hyun <dhyun@apple.com> Signed-off-by: Dongjoon Hyun <dhyun@apple.com>
2019-10-03 11:58:25 -04:00
date_sub: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative
------------------------------------------------------------------------------------------------------------------------
[SPARK-31630][SQL] Fix perf regression by skipping timestamps rebasing after some threshold ### What changes were proposed in this pull request? Skip timestamps rebasing after a global threshold when there is no difference between Julian and Gregorian calendars. This allows to avoid checking hash maps of switch points, and fixes perf regressions in `toJavaTimestamp()` and `fromJavaTimestamp()`. ### Why are the changes needed? The changes fix perf regressions of conversions to/from external type `java.sql.Timestamp`. Before (see the PR's results https://github.com/apache/spark/pull/28440): ``` ================================================================================================ Conversion from/to external types ================================================================================================ OpenJDK 64-Bit Server VM 1.8.0_252-8u252-b09-1~18.04-b09 on Linux 4.15.0-1063-aws Intel(R) Xeon(R) CPU E5-2670 v2 2.50GHz To/from Java's date-time: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative ------------------------------------------------------------------------------------------------------------------------ From java.sql.Timestamp 376 388 10 13.3 75.2 1.1X Collect java.sql.Timestamp 1878 1937 64 2.7 375.6 0.2X ``` After: ``` ================================================================================================ Conversion from/to external types ================================================================================================ OpenJDK 64-Bit Server VM 1.8.0_252-8u252-b09-1~18.04-b09 on Linux 4.15.0-1063-aws Intel(R) Xeon(R) CPU E5-2670 v2 2.50GHz To/from Java's date-time: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative ------------------------------------------------------------------------------------------------------------------------ From java.sql.Timestamp 249 264 24 20.1 49.8 1.7X Collect java.sql.Timestamp 1503 1523 24 3.3 300.5 0.3X ``` Perf improvements in average of: 1. From java.sql.Timestamp is ~ 34% 2. To java.sql.Timestamps is ~16% ### Does this PR introduce _any_ user-facing change? No ### How was this patch tested? By existing test suites `DateTimeUtilsSuite` and `RebaseDateTimeSuite`. Closes #28441 from MaxGekk/opt-rebase-common-threshold. Authored-by: Max Gekk <max.gekk@gmail.com> Signed-off-by: Wenchen Fan <wenchen@databricks.com>
2020-05-05 10:11:53 -04:00
date_sub wholestage off 1082 1083 1 9.2 108.2 1.0X
date_sub wholestage on 1047 1056 12 9.6 104.7 1.0X
[SPARK-29320][TESTS] Compare `sql/core` module in JDK8/11 (Part 1) ### What changes were proposed in this pull request? This PR regenerates the `sql/core` benchmarks in JDK8/11 to compare the result. In general, we compare the ratio instead of the time. However, in this PR, the average time is compared. This PR should be considered as a rough comparison. **A. EXPECTED CASES(JDK11 is faster in general)** - [x] BloomFilterBenchmark (JDK11 is faster except one case) - [x] BuiltInDataSourceWriteBenchmark (JDK11 is faster at CSV/ORC) - [x] CSVBenchmark (JDK11 is faster except five cases) - [x] ColumnarBatchBenchmark (JDK11 is faster at `boolean`/`string` and some cases in `int`/`array`) - [x] DatasetBenchmark (JDK11 is faster with `string`, but is slower for `long` type) - [x] ExternalAppendOnlyUnsafeRowArrayBenchmark (JDK11 is faster except two cases) - [x] ExtractBenchmark (JDK11 is faster except HOUR/MINUTE/SECOND/MILLISECONDS/MICROSECONDS) - [x] HashedRelationMetricsBenchmark (JDK11 is faster) - [x] JSONBenchmark (JDK11 is much faster except eight cases) - [x] JoinBenchmark (JDK11 is faster except five cases) - [x] OrcNestedSchemaPruningBenchmark (JDK11 is faster in nine cases) - [x] PrimitiveArrayBenchmark (JDK11 is faster) - [x] SortBenchmark (JDK11 is faster except `Arrays.sort` case) - [x] UDFBenchmark (N/A, values are too small) - [x] UnsafeArrayDataBenchmark (JDK11 is faster except one case) - [x] WideTableBenchmark (JDK11 is faster except two cases) **B. CASES WE NEED TO INVESTIGATE MORE LATER** - [x] AggregateBenchmark (JDK11 is slower in general) - [x] CompressionSchemeBenchmark (JDK11 is slower in general except `string`) - [x] DataSourceReadBenchmark (JDK11 is slower in general) - [x] DateTimeBenchmark (JDK11 is slightly slower in general except `parsing`) - [x] MakeDateTimeBenchmark (JDK11 is slower except two cases) - [x] MiscBenchmark (JDK11 is slower except ten cases) - [x] OrcV2NestedSchemaPruningBenchmark (JDK11 is slower) - [x] ParquetNestedSchemaPruningBenchmark (JDK11 is slower except six cases) - [x] RangeBenchmark (JDK11 is slower except one case) `FilterPushdownBenchmark/InExpressionBenchmark/WideSchemaBenchmark` will be compared later because it took long timer. ### Why are the changes needed? According to the result, there are some difference between JDK8/JDK11. This will be a baseline for the future improvement and comparison. Also, as a reproducible environment, the following environment is used. - Instance: `r3.xlarge` - OS: `CentOS Linux release 7.5.1804 (Core)` - JDK: - `OpenJDK Runtime Environment (build 1.8.0_222-b10)` - `OpenJDK Runtime Environment 18.9 (build 11.0.4+11-LTS)` ### Does this PR introduce any user-facing change? No. ### How was this patch tested? This is a test-only PR. We need to run benchmark. Closes #26003 from dongjoon-hyun/SPARK-29320. Authored-by: Dongjoon Hyun <dhyun@apple.com> Signed-off-by: Dongjoon Hyun <dhyun@apple.com>
2019-10-03 11:58:25 -04:00
[SPARK-31527][SQL][TESTS][FOLLOWUP] Fix the number of rows in `DateTimeBenchmark` ### What changes were proposed in this pull request? - Changed to the number of rows in benchmark cases from 3 to the actual number `N`. - Regenerated benchmark results in the environment: | Item | Description | | ---- | ----| | Region | us-west-2 (Oregon) | | Instance | r3.xlarge | | AMI | ubuntu/images/hvm-ssd/ubuntu-bionic-18.04-amd64-server-20190722.1 (ami-06f2f779464715dc5) | | Java | OpenJDK 64-Bit Server VM 1.8.0_242 and OpenJDK 64-Bit Server VM 11.0.6+10 | ### Why are the changes needed? The changes are needed to have: - Correct benchmark results - Base line for other perf improvements that can be checked in the same environment. ### Does this PR introduce _any_ user-facing change? No ### How was this patch tested? By running the benchmark and checking its output. Closes #28440 from MaxGekk/SPARK-31527-DateTimeBenchmark-followup. Authored-by: Max Gekk <max.gekk@gmail.com> Signed-off-by: Takeshi Yamamuro <yamamuro@apache.org>
2020-05-03 20:39:50 -04:00
OpenJDK 64-Bit Server VM 11.0.7+10-post-Ubuntu-2ubuntu218.04 on Linux 4.15.0-1063-aws
Intel(R) Xeon(R) CPU E5-2670 v2 @ 2.50GHz
[SPARK-29320][TESTS] Compare `sql/core` module in JDK8/11 (Part 1) ### What changes were proposed in this pull request? This PR regenerates the `sql/core` benchmarks in JDK8/11 to compare the result. In general, we compare the ratio instead of the time. However, in this PR, the average time is compared. This PR should be considered as a rough comparison. **A. EXPECTED CASES(JDK11 is faster in general)** - [x] BloomFilterBenchmark (JDK11 is faster except one case) - [x] BuiltInDataSourceWriteBenchmark (JDK11 is faster at CSV/ORC) - [x] CSVBenchmark (JDK11 is faster except five cases) - [x] ColumnarBatchBenchmark (JDK11 is faster at `boolean`/`string` and some cases in `int`/`array`) - [x] DatasetBenchmark (JDK11 is faster with `string`, but is slower for `long` type) - [x] ExternalAppendOnlyUnsafeRowArrayBenchmark (JDK11 is faster except two cases) - [x] ExtractBenchmark (JDK11 is faster except HOUR/MINUTE/SECOND/MILLISECONDS/MICROSECONDS) - [x] HashedRelationMetricsBenchmark (JDK11 is faster) - [x] JSONBenchmark (JDK11 is much faster except eight cases) - [x] JoinBenchmark (JDK11 is faster except five cases) - [x] OrcNestedSchemaPruningBenchmark (JDK11 is faster in nine cases) - [x] PrimitiveArrayBenchmark (JDK11 is faster) - [x] SortBenchmark (JDK11 is faster except `Arrays.sort` case) - [x] UDFBenchmark (N/A, values are too small) - [x] UnsafeArrayDataBenchmark (JDK11 is faster except one case) - [x] WideTableBenchmark (JDK11 is faster except two cases) **B. CASES WE NEED TO INVESTIGATE MORE LATER** - [x] AggregateBenchmark (JDK11 is slower in general) - [x] CompressionSchemeBenchmark (JDK11 is slower in general except `string`) - [x] DataSourceReadBenchmark (JDK11 is slower in general) - [x] DateTimeBenchmark (JDK11 is slightly slower in general except `parsing`) - [x] MakeDateTimeBenchmark (JDK11 is slower except two cases) - [x] MiscBenchmark (JDK11 is slower except ten cases) - [x] OrcV2NestedSchemaPruningBenchmark (JDK11 is slower) - [x] ParquetNestedSchemaPruningBenchmark (JDK11 is slower except six cases) - [x] RangeBenchmark (JDK11 is slower except one case) `FilterPushdownBenchmark/InExpressionBenchmark/WideSchemaBenchmark` will be compared later because it took long timer. ### Why are the changes needed? According to the result, there are some difference between JDK8/JDK11. This will be a baseline for the future improvement and comparison. Also, as a reproducible environment, the following environment is used. - Instance: `r3.xlarge` - OS: `CentOS Linux release 7.5.1804 (Core)` - JDK: - `OpenJDK Runtime Environment (build 1.8.0_222-b10)` - `OpenJDK Runtime Environment 18.9 (build 11.0.4+11-LTS)` ### Does this PR introduce any user-facing change? No. ### How was this patch tested? This is a test-only PR. We need to run benchmark. Closes #26003 from dongjoon-hyun/SPARK-29320. Authored-by: Dongjoon Hyun <dhyun@apple.com> Signed-off-by: Dongjoon Hyun <dhyun@apple.com>
2019-10-03 11:58:25 -04:00
add_months: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative
------------------------------------------------------------------------------------------------------------------------
[SPARK-31630][SQL] Fix perf regression by skipping timestamps rebasing after some threshold ### What changes were proposed in this pull request? Skip timestamps rebasing after a global threshold when there is no difference between Julian and Gregorian calendars. This allows to avoid checking hash maps of switch points, and fixes perf regressions in `toJavaTimestamp()` and `fromJavaTimestamp()`. ### Why are the changes needed? The changes fix perf regressions of conversions to/from external type `java.sql.Timestamp`. Before (see the PR's results https://github.com/apache/spark/pull/28440): ``` ================================================================================================ Conversion from/to external types ================================================================================================ OpenJDK 64-Bit Server VM 1.8.0_252-8u252-b09-1~18.04-b09 on Linux 4.15.0-1063-aws Intel(R) Xeon(R) CPU E5-2670 v2 2.50GHz To/from Java's date-time: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative ------------------------------------------------------------------------------------------------------------------------ From java.sql.Timestamp 376 388 10 13.3 75.2 1.1X Collect java.sql.Timestamp 1878 1937 64 2.7 375.6 0.2X ``` After: ``` ================================================================================================ Conversion from/to external types ================================================================================================ OpenJDK 64-Bit Server VM 1.8.0_252-8u252-b09-1~18.04-b09 on Linux 4.15.0-1063-aws Intel(R) Xeon(R) CPU E5-2670 v2 2.50GHz To/from Java's date-time: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative ------------------------------------------------------------------------------------------------------------------------ From java.sql.Timestamp 249 264 24 20.1 49.8 1.7X Collect java.sql.Timestamp 1503 1523 24 3.3 300.5 0.3X ``` Perf improvements in average of: 1. From java.sql.Timestamp is ~ 34% 2. To java.sql.Timestamps is ~16% ### Does this PR introduce _any_ user-facing change? No ### How was this patch tested? By existing test suites `DateTimeUtilsSuite` and `RebaseDateTimeSuite`. Closes #28441 from MaxGekk/opt-rebase-common-threshold. Authored-by: Max Gekk <max.gekk@gmail.com> Signed-off-by: Wenchen Fan <wenchen@databricks.com>
2020-05-05 10:11:53 -04:00
add_months wholestage off 1430 1431 1 7.0 143.0 1.0X
add_months wholestage on 1441 1446 8 6.9 144.1 1.0X
[SPARK-29320][TESTS] Compare `sql/core` module in JDK8/11 (Part 1) ### What changes were proposed in this pull request? This PR regenerates the `sql/core` benchmarks in JDK8/11 to compare the result. In general, we compare the ratio instead of the time. However, in this PR, the average time is compared. This PR should be considered as a rough comparison. **A. EXPECTED CASES(JDK11 is faster in general)** - [x] BloomFilterBenchmark (JDK11 is faster except one case) - [x] BuiltInDataSourceWriteBenchmark (JDK11 is faster at CSV/ORC) - [x] CSVBenchmark (JDK11 is faster except five cases) - [x] ColumnarBatchBenchmark (JDK11 is faster at `boolean`/`string` and some cases in `int`/`array`) - [x] DatasetBenchmark (JDK11 is faster with `string`, but is slower for `long` type) - [x] ExternalAppendOnlyUnsafeRowArrayBenchmark (JDK11 is faster except two cases) - [x] ExtractBenchmark (JDK11 is faster except HOUR/MINUTE/SECOND/MILLISECONDS/MICROSECONDS) - [x] HashedRelationMetricsBenchmark (JDK11 is faster) - [x] JSONBenchmark (JDK11 is much faster except eight cases) - [x] JoinBenchmark (JDK11 is faster except five cases) - [x] OrcNestedSchemaPruningBenchmark (JDK11 is faster in nine cases) - [x] PrimitiveArrayBenchmark (JDK11 is faster) - [x] SortBenchmark (JDK11 is faster except `Arrays.sort` case) - [x] UDFBenchmark (N/A, values are too small) - [x] UnsafeArrayDataBenchmark (JDK11 is faster except one case) - [x] WideTableBenchmark (JDK11 is faster except two cases) **B. CASES WE NEED TO INVESTIGATE MORE LATER** - [x] AggregateBenchmark (JDK11 is slower in general) - [x] CompressionSchemeBenchmark (JDK11 is slower in general except `string`) - [x] DataSourceReadBenchmark (JDK11 is slower in general) - [x] DateTimeBenchmark (JDK11 is slightly slower in general except `parsing`) - [x] MakeDateTimeBenchmark (JDK11 is slower except two cases) - [x] MiscBenchmark (JDK11 is slower except ten cases) - [x] OrcV2NestedSchemaPruningBenchmark (JDK11 is slower) - [x] ParquetNestedSchemaPruningBenchmark (JDK11 is slower except six cases) - [x] RangeBenchmark (JDK11 is slower except one case) `FilterPushdownBenchmark/InExpressionBenchmark/WideSchemaBenchmark` will be compared later because it took long timer. ### Why are the changes needed? According to the result, there are some difference between JDK8/JDK11. This will be a baseline for the future improvement and comparison. Also, as a reproducible environment, the following environment is used. - Instance: `r3.xlarge` - OS: `CentOS Linux release 7.5.1804 (Core)` - JDK: - `OpenJDK Runtime Environment (build 1.8.0_222-b10)` - `OpenJDK Runtime Environment 18.9 (build 11.0.4+11-LTS)` ### Does this PR introduce any user-facing change? No. ### How was this patch tested? This is a test-only PR. We need to run benchmark. Closes #26003 from dongjoon-hyun/SPARK-29320. Authored-by: Dongjoon Hyun <dhyun@apple.com> Signed-off-by: Dongjoon Hyun <dhyun@apple.com>
2019-10-03 11:58:25 -04:00
================================================================================================
Formatting dates
================================================================================================
[SPARK-31527][SQL][TESTS][FOLLOWUP] Fix the number of rows in `DateTimeBenchmark` ### What changes were proposed in this pull request? - Changed to the number of rows in benchmark cases from 3 to the actual number `N`. - Regenerated benchmark results in the environment: | Item | Description | | ---- | ----| | Region | us-west-2 (Oregon) | | Instance | r3.xlarge | | AMI | ubuntu/images/hvm-ssd/ubuntu-bionic-18.04-amd64-server-20190722.1 (ami-06f2f779464715dc5) | | Java | OpenJDK 64-Bit Server VM 1.8.0_242 and OpenJDK 64-Bit Server VM 11.0.6+10 | ### Why are the changes needed? The changes are needed to have: - Correct benchmark results - Base line for other perf improvements that can be checked in the same environment. ### Does this PR introduce _any_ user-facing change? No ### How was this patch tested? By running the benchmark and checking its output. Closes #28440 from MaxGekk/SPARK-31527-DateTimeBenchmark-followup. Authored-by: Max Gekk <max.gekk@gmail.com> Signed-off-by: Takeshi Yamamuro <yamamuro@apache.org>
2020-05-03 20:39:50 -04:00
OpenJDK 64-Bit Server VM 11.0.7+10-post-Ubuntu-2ubuntu218.04 on Linux 4.15.0-1063-aws
Intel(R) Xeon(R) CPU E5-2670 v2 @ 2.50GHz
[SPARK-29320][TESTS] Compare `sql/core` module in JDK8/11 (Part 1) ### What changes were proposed in this pull request? This PR regenerates the `sql/core` benchmarks in JDK8/11 to compare the result. In general, we compare the ratio instead of the time. However, in this PR, the average time is compared. This PR should be considered as a rough comparison. **A. EXPECTED CASES(JDK11 is faster in general)** - [x] BloomFilterBenchmark (JDK11 is faster except one case) - [x] BuiltInDataSourceWriteBenchmark (JDK11 is faster at CSV/ORC) - [x] CSVBenchmark (JDK11 is faster except five cases) - [x] ColumnarBatchBenchmark (JDK11 is faster at `boolean`/`string` and some cases in `int`/`array`) - [x] DatasetBenchmark (JDK11 is faster with `string`, but is slower for `long` type) - [x] ExternalAppendOnlyUnsafeRowArrayBenchmark (JDK11 is faster except two cases) - [x] ExtractBenchmark (JDK11 is faster except HOUR/MINUTE/SECOND/MILLISECONDS/MICROSECONDS) - [x] HashedRelationMetricsBenchmark (JDK11 is faster) - [x] JSONBenchmark (JDK11 is much faster except eight cases) - [x] JoinBenchmark (JDK11 is faster except five cases) - [x] OrcNestedSchemaPruningBenchmark (JDK11 is faster in nine cases) - [x] PrimitiveArrayBenchmark (JDK11 is faster) - [x] SortBenchmark (JDK11 is faster except `Arrays.sort` case) - [x] UDFBenchmark (N/A, values are too small) - [x] UnsafeArrayDataBenchmark (JDK11 is faster except one case) - [x] WideTableBenchmark (JDK11 is faster except two cases) **B. CASES WE NEED TO INVESTIGATE MORE LATER** - [x] AggregateBenchmark (JDK11 is slower in general) - [x] CompressionSchemeBenchmark (JDK11 is slower in general except `string`) - [x] DataSourceReadBenchmark (JDK11 is slower in general) - [x] DateTimeBenchmark (JDK11 is slightly slower in general except `parsing`) - [x] MakeDateTimeBenchmark (JDK11 is slower except two cases) - [x] MiscBenchmark (JDK11 is slower except ten cases) - [x] OrcV2NestedSchemaPruningBenchmark (JDK11 is slower) - [x] ParquetNestedSchemaPruningBenchmark (JDK11 is slower except six cases) - [x] RangeBenchmark (JDK11 is slower except one case) `FilterPushdownBenchmark/InExpressionBenchmark/WideSchemaBenchmark` will be compared later because it took long timer. ### Why are the changes needed? According to the result, there are some difference between JDK8/JDK11. This will be a baseline for the future improvement and comparison. Also, as a reproducible environment, the following environment is used. - Instance: `r3.xlarge` - OS: `CentOS Linux release 7.5.1804 (Core)` - JDK: - `OpenJDK Runtime Environment (build 1.8.0_222-b10)` - `OpenJDK Runtime Environment 18.9 (build 11.0.4+11-LTS)` ### Does this PR introduce any user-facing change? No. ### How was this patch tested? This is a test-only PR. We need to run benchmark. Closes #26003 from dongjoon-hyun/SPARK-29320. Authored-by: Dongjoon Hyun <dhyun@apple.com> Signed-off-by: Dongjoon Hyun <dhyun@apple.com>
2019-10-03 11:58:25 -04:00
format date: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative
------------------------------------------------------------------------------------------------------------------------
[SPARK-31630][SQL] Fix perf regression by skipping timestamps rebasing after some threshold ### What changes were proposed in this pull request? Skip timestamps rebasing after a global threshold when there is no difference between Julian and Gregorian calendars. This allows to avoid checking hash maps of switch points, and fixes perf regressions in `toJavaTimestamp()` and `fromJavaTimestamp()`. ### Why are the changes needed? The changes fix perf regressions of conversions to/from external type `java.sql.Timestamp`. Before (see the PR's results https://github.com/apache/spark/pull/28440): ``` ================================================================================================ Conversion from/to external types ================================================================================================ OpenJDK 64-Bit Server VM 1.8.0_252-8u252-b09-1~18.04-b09 on Linux 4.15.0-1063-aws Intel(R) Xeon(R) CPU E5-2670 v2 2.50GHz To/from Java's date-time: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative ------------------------------------------------------------------------------------------------------------------------ From java.sql.Timestamp 376 388 10 13.3 75.2 1.1X Collect java.sql.Timestamp 1878 1937 64 2.7 375.6 0.2X ``` After: ``` ================================================================================================ Conversion from/to external types ================================================================================================ OpenJDK 64-Bit Server VM 1.8.0_252-8u252-b09-1~18.04-b09 on Linux 4.15.0-1063-aws Intel(R) Xeon(R) CPU E5-2670 v2 2.50GHz To/from Java's date-time: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative ------------------------------------------------------------------------------------------------------------------------ From java.sql.Timestamp 249 264 24 20.1 49.8 1.7X Collect java.sql.Timestamp 1503 1523 24 3.3 300.5 0.3X ``` Perf improvements in average of: 1. From java.sql.Timestamp is ~ 34% 2. To java.sql.Timestamps is ~16% ### Does this PR introduce _any_ user-facing change? No ### How was this patch tested? By existing test suites `DateTimeUtilsSuite` and `RebaseDateTimeSuite`. Closes #28441 from MaxGekk/opt-rebase-common-threshold. Authored-by: Max Gekk <max.gekk@gmail.com> Signed-off-by: Wenchen Fan <wenchen@databricks.com>
2020-05-05 10:11:53 -04:00
format date wholestage off 5442 5549 150 1.8 544.2 1.0X
format date wholestage on 5529 5655 236 1.8 552.9 1.0X
[SPARK-29320][TESTS] Compare `sql/core` module in JDK8/11 (Part 1) ### What changes were proposed in this pull request? This PR regenerates the `sql/core` benchmarks in JDK8/11 to compare the result. In general, we compare the ratio instead of the time. However, in this PR, the average time is compared. This PR should be considered as a rough comparison. **A. EXPECTED CASES(JDK11 is faster in general)** - [x] BloomFilterBenchmark (JDK11 is faster except one case) - [x] BuiltInDataSourceWriteBenchmark (JDK11 is faster at CSV/ORC) - [x] CSVBenchmark (JDK11 is faster except five cases) - [x] ColumnarBatchBenchmark (JDK11 is faster at `boolean`/`string` and some cases in `int`/`array`) - [x] DatasetBenchmark (JDK11 is faster with `string`, but is slower for `long` type) - [x] ExternalAppendOnlyUnsafeRowArrayBenchmark (JDK11 is faster except two cases) - [x] ExtractBenchmark (JDK11 is faster except HOUR/MINUTE/SECOND/MILLISECONDS/MICROSECONDS) - [x] HashedRelationMetricsBenchmark (JDK11 is faster) - [x] JSONBenchmark (JDK11 is much faster except eight cases) - [x] JoinBenchmark (JDK11 is faster except five cases) - [x] OrcNestedSchemaPruningBenchmark (JDK11 is faster in nine cases) - [x] PrimitiveArrayBenchmark (JDK11 is faster) - [x] SortBenchmark (JDK11 is faster except `Arrays.sort` case) - [x] UDFBenchmark (N/A, values are too small) - [x] UnsafeArrayDataBenchmark (JDK11 is faster except one case) - [x] WideTableBenchmark (JDK11 is faster except two cases) **B. CASES WE NEED TO INVESTIGATE MORE LATER** - [x] AggregateBenchmark (JDK11 is slower in general) - [x] CompressionSchemeBenchmark (JDK11 is slower in general except `string`) - [x] DataSourceReadBenchmark (JDK11 is slower in general) - [x] DateTimeBenchmark (JDK11 is slightly slower in general except `parsing`) - [x] MakeDateTimeBenchmark (JDK11 is slower except two cases) - [x] MiscBenchmark (JDK11 is slower except ten cases) - [x] OrcV2NestedSchemaPruningBenchmark (JDK11 is slower) - [x] ParquetNestedSchemaPruningBenchmark (JDK11 is slower except six cases) - [x] RangeBenchmark (JDK11 is slower except one case) `FilterPushdownBenchmark/InExpressionBenchmark/WideSchemaBenchmark` will be compared later because it took long timer. ### Why are the changes needed? According to the result, there are some difference between JDK8/JDK11. This will be a baseline for the future improvement and comparison. Also, as a reproducible environment, the following environment is used. - Instance: `r3.xlarge` - OS: `CentOS Linux release 7.5.1804 (Core)` - JDK: - `OpenJDK Runtime Environment (build 1.8.0_222-b10)` - `OpenJDK Runtime Environment 18.9 (build 11.0.4+11-LTS)` ### Does this PR introduce any user-facing change? No. ### How was this patch tested? This is a test-only PR. We need to run benchmark. Closes #26003 from dongjoon-hyun/SPARK-29320. Authored-by: Dongjoon Hyun <dhyun@apple.com> Signed-off-by: Dongjoon Hyun <dhyun@apple.com>
2019-10-03 11:58:25 -04:00
================================================================================================
Formatting timestamps
================================================================================================
[SPARK-31527][SQL][TESTS][FOLLOWUP] Fix the number of rows in `DateTimeBenchmark` ### What changes were proposed in this pull request? - Changed to the number of rows in benchmark cases from 3 to the actual number `N`. - Regenerated benchmark results in the environment: | Item | Description | | ---- | ----| | Region | us-west-2 (Oregon) | | Instance | r3.xlarge | | AMI | ubuntu/images/hvm-ssd/ubuntu-bionic-18.04-amd64-server-20190722.1 (ami-06f2f779464715dc5) | | Java | OpenJDK 64-Bit Server VM 1.8.0_242 and OpenJDK 64-Bit Server VM 11.0.6+10 | ### Why are the changes needed? The changes are needed to have: - Correct benchmark results - Base line for other perf improvements that can be checked in the same environment. ### Does this PR introduce _any_ user-facing change? No ### How was this patch tested? By running the benchmark and checking its output. Closes #28440 from MaxGekk/SPARK-31527-DateTimeBenchmark-followup. Authored-by: Max Gekk <max.gekk@gmail.com> Signed-off-by: Takeshi Yamamuro <yamamuro@apache.org>
2020-05-03 20:39:50 -04:00
OpenJDK 64-Bit Server VM 11.0.7+10-post-Ubuntu-2ubuntu218.04 on Linux 4.15.0-1063-aws
Intel(R) Xeon(R) CPU E5-2670 v2 @ 2.50GHz
[SPARK-29320][TESTS] Compare `sql/core` module in JDK8/11 (Part 1) ### What changes were proposed in this pull request? This PR regenerates the `sql/core` benchmarks in JDK8/11 to compare the result. In general, we compare the ratio instead of the time. However, in this PR, the average time is compared. This PR should be considered as a rough comparison. **A. EXPECTED CASES(JDK11 is faster in general)** - [x] BloomFilterBenchmark (JDK11 is faster except one case) - [x] BuiltInDataSourceWriteBenchmark (JDK11 is faster at CSV/ORC) - [x] CSVBenchmark (JDK11 is faster except five cases) - [x] ColumnarBatchBenchmark (JDK11 is faster at `boolean`/`string` and some cases in `int`/`array`) - [x] DatasetBenchmark (JDK11 is faster with `string`, but is slower for `long` type) - [x] ExternalAppendOnlyUnsafeRowArrayBenchmark (JDK11 is faster except two cases) - [x] ExtractBenchmark (JDK11 is faster except HOUR/MINUTE/SECOND/MILLISECONDS/MICROSECONDS) - [x] HashedRelationMetricsBenchmark (JDK11 is faster) - [x] JSONBenchmark (JDK11 is much faster except eight cases) - [x] JoinBenchmark (JDK11 is faster except five cases) - [x] OrcNestedSchemaPruningBenchmark (JDK11 is faster in nine cases) - [x] PrimitiveArrayBenchmark (JDK11 is faster) - [x] SortBenchmark (JDK11 is faster except `Arrays.sort` case) - [x] UDFBenchmark (N/A, values are too small) - [x] UnsafeArrayDataBenchmark (JDK11 is faster except one case) - [x] WideTableBenchmark (JDK11 is faster except two cases) **B. CASES WE NEED TO INVESTIGATE MORE LATER** - [x] AggregateBenchmark (JDK11 is slower in general) - [x] CompressionSchemeBenchmark (JDK11 is slower in general except `string`) - [x] DataSourceReadBenchmark (JDK11 is slower in general) - [x] DateTimeBenchmark (JDK11 is slightly slower in general except `parsing`) - [x] MakeDateTimeBenchmark (JDK11 is slower except two cases) - [x] MiscBenchmark (JDK11 is slower except ten cases) - [x] OrcV2NestedSchemaPruningBenchmark (JDK11 is slower) - [x] ParquetNestedSchemaPruningBenchmark (JDK11 is slower except six cases) - [x] RangeBenchmark (JDK11 is slower except one case) `FilterPushdownBenchmark/InExpressionBenchmark/WideSchemaBenchmark` will be compared later because it took long timer. ### Why are the changes needed? According to the result, there are some difference between JDK8/JDK11. This will be a baseline for the future improvement and comparison. Also, as a reproducible environment, the following environment is used. - Instance: `r3.xlarge` - OS: `CentOS Linux release 7.5.1804 (Core)` - JDK: - `OpenJDK Runtime Environment (build 1.8.0_222-b10)` - `OpenJDK Runtime Environment 18.9 (build 11.0.4+11-LTS)` ### Does this PR introduce any user-facing change? No. ### How was this patch tested? This is a test-only PR. We need to run benchmark. Closes #26003 from dongjoon-hyun/SPARK-29320. Authored-by: Dongjoon Hyun <dhyun@apple.com> Signed-off-by: Dongjoon Hyun <dhyun@apple.com>
2019-10-03 11:58:25 -04:00
from_unixtime: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative
------------------------------------------------------------------------------------------------------------------------
[SPARK-31630][SQL] Fix perf regression by skipping timestamps rebasing after some threshold ### What changes were proposed in this pull request? Skip timestamps rebasing after a global threshold when there is no difference between Julian and Gregorian calendars. This allows to avoid checking hash maps of switch points, and fixes perf regressions in `toJavaTimestamp()` and `fromJavaTimestamp()`. ### Why are the changes needed? The changes fix perf regressions of conversions to/from external type `java.sql.Timestamp`. Before (see the PR's results https://github.com/apache/spark/pull/28440): ``` ================================================================================================ Conversion from/to external types ================================================================================================ OpenJDK 64-Bit Server VM 1.8.0_252-8u252-b09-1~18.04-b09 on Linux 4.15.0-1063-aws Intel(R) Xeon(R) CPU E5-2670 v2 2.50GHz To/from Java's date-time: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative ------------------------------------------------------------------------------------------------------------------------ From java.sql.Timestamp 376 388 10 13.3 75.2 1.1X Collect java.sql.Timestamp 1878 1937 64 2.7 375.6 0.2X ``` After: ``` ================================================================================================ Conversion from/to external types ================================================================================================ OpenJDK 64-Bit Server VM 1.8.0_252-8u252-b09-1~18.04-b09 on Linux 4.15.0-1063-aws Intel(R) Xeon(R) CPU E5-2670 v2 2.50GHz To/from Java's date-time: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative ------------------------------------------------------------------------------------------------------------------------ From java.sql.Timestamp 249 264 24 20.1 49.8 1.7X Collect java.sql.Timestamp 1503 1523 24 3.3 300.5 0.3X ``` Perf improvements in average of: 1. From java.sql.Timestamp is ~ 34% 2. To java.sql.Timestamps is ~16% ### Does this PR introduce _any_ user-facing change? No ### How was this patch tested? By existing test suites `DateTimeUtilsSuite` and `RebaseDateTimeSuite`. Closes #28441 from MaxGekk/opt-rebase-common-threshold. Authored-by: Max Gekk <max.gekk@gmail.com> Signed-off-by: Wenchen Fan <wenchen@databricks.com>
2020-05-05 10:11:53 -04:00
from_unixtime wholestage off 7416 7440 34 1.3 741.6 1.0X
from_unixtime wholestage on 7372 7391 17 1.4 737.2 1.0X
[SPARK-29320][TESTS] Compare `sql/core` module in JDK8/11 (Part 1) ### What changes were proposed in this pull request? This PR regenerates the `sql/core` benchmarks in JDK8/11 to compare the result. In general, we compare the ratio instead of the time. However, in this PR, the average time is compared. This PR should be considered as a rough comparison. **A. EXPECTED CASES(JDK11 is faster in general)** - [x] BloomFilterBenchmark (JDK11 is faster except one case) - [x] BuiltInDataSourceWriteBenchmark (JDK11 is faster at CSV/ORC) - [x] CSVBenchmark (JDK11 is faster except five cases) - [x] ColumnarBatchBenchmark (JDK11 is faster at `boolean`/`string` and some cases in `int`/`array`) - [x] DatasetBenchmark (JDK11 is faster with `string`, but is slower for `long` type) - [x] ExternalAppendOnlyUnsafeRowArrayBenchmark (JDK11 is faster except two cases) - [x] ExtractBenchmark (JDK11 is faster except HOUR/MINUTE/SECOND/MILLISECONDS/MICROSECONDS) - [x] HashedRelationMetricsBenchmark (JDK11 is faster) - [x] JSONBenchmark (JDK11 is much faster except eight cases) - [x] JoinBenchmark (JDK11 is faster except five cases) - [x] OrcNestedSchemaPruningBenchmark (JDK11 is faster in nine cases) - [x] PrimitiveArrayBenchmark (JDK11 is faster) - [x] SortBenchmark (JDK11 is faster except `Arrays.sort` case) - [x] UDFBenchmark (N/A, values are too small) - [x] UnsafeArrayDataBenchmark (JDK11 is faster except one case) - [x] WideTableBenchmark (JDK11 is faster except two cases) **B. CASES WE NEED TO INVESTIGATE MORE LATER** - [x] AggregateBenchmark (JDK11 is slower in general) - [x] CompressionSchemeBenchmark (JDK11 is slower in general except `string`) - [x] DataSourceReadBenchmark (JDK11 is slower in general) - [x] DateTimeBenchmark (JDK11 is slightly slower in general except `parsing`) - [x] MakeDateTimeBenchmark (JDK11 is slower except two cases) - [x] MiscBenchmark (JDK11 is slower except ten cases) - [x] OrcV2NestedSchemaPruningBenchmark (JDK11 is slower) - [x] ParquetNestedSchemaPruningBenchmark (JDK11 is slower except six cases) - [x] RangeBenchmark (JDK11 is slower except one case) `FilterPushdownBenchmark/InExpressionBenchmark/WideSchemaBenchmark` will be compared later because it took long timer. ### Why are the changes needed? According to the result, there are some difference between JDK8/JDK11. This will be a baseline for the future improvement and comparison. Also, as a reproducible environment, the following environment is used. - Instance: `r3.xlarge` - OS: `CentOS Linux release 7.5.1804 (Core)` - JDK: - `OpenJDK Runtime Environment (build 1.8.0_222-b10)` - `OpenJDK Runtime Environment 18.9 (build 11.0.4+11-LTS)` ### Does this PR introduce any user-facing change? No. ### How was this patch tested? This is a test-only PR. We need to run benchmark. Closes #26003 from dongjoon-hyun/SPARK-29320. Authored-by: Dongjoon Hyun <dhyun@apple.com> Signed-off-by: Dongjoon Hyun <dhyun@apple.com>
2019-10-03 11:58:25 -04:00
================================================================================================
Convert timestamps
================================================================================================
[SPARK-31527][SQL][TESTS][FOLLOWUP] Fix the number of rows in `DateTimeBenchmark` ### What changes were proposed in this pull request? - Changed to the number of rows in benchmark cases from 3 to the actual number `N`. - Regenerated benchmark results in the environment: | Item | Description | | ---- | ----| | Region | us-west-2 (Oregon) | | Instance | r3.xlarge | | AMI | ubuntu/images/hvm-ssd/ubuntu-bionic-18.04-amd64-server-20190722.1 (ami-06f2f779464715dc5) | | Java | OpenJDK 64-Bit Server VM 1.8.0_242 and OpenJDK 64-Bit Server VM 11.0.6+10 | ### Why are the changes needed? The changes are needed to have: - Correct benchmark results - Base line for other perf improvements that can be checked in the same environment. ### Does this PR introduce _any_ user-facing change? No ### How was this patch tested? By running the benchmark and checking its output. Closes #28440 from MaxGekk/SPARK-31527-DateTimeBenchmark-followup. Authored-by: Max Gekk <max.gekk@gmail.com> Signed-off-by: Takeshi Yamamuro <yamamuro@apache.org>
2020-05-03 20:39:50 -04:00
OpenJDK 64-Bit Server VM 11.0.7+10-post-Ubuntu-2ubuntu218.04 on Linux 4.15.0-1063-aws
Intel(R) Xeon(R) CPU E5-2670 v2 @ 2.50GHz
[SPARK-29320][TESTS] Compare `sql/core` module in JDK8/11 (Part 1) ### What changes were proposed in this pull request? This PR regenerates the `sql/core` benchmarks in JDK8/11 to compare the result. In general, we compare the ratio instead of the time. However, in this PR, the average time is compared. This PR should be considered as a rough comparison. **A. EXPECTED CASES(JDK11 is faster in general)** - [x] BloomFilterBenchmark (JDK11 is faster except one case) - [x] BuiltInDataSourceWriteBenchmark (JDK11 is faster at CSV/ORC) - [x] CSVBenchmark (JDK11 is faster except five cases) - [x] ColumnarBatchBenchmark (JDK11 is faster at `boolean`/`string` and some cases in `int`/`array`) - [x] DatasetBenchmark (JDK11 is faster with `string`, but is slower for `long` type) - [x] ExternalAppendOnlyUnsafeRowArrayBenchmark (JDK11 is faster except two cases) - [x] ExtractBenchmark (JDK11 is faster except HOUR/MINUTE/SECOND/MILLISECONDS/MICROSECONDS) - [x] HashedRelationMetricsBenchmark (JDK11 is faster) - [x] JSONBenchmark (JDK11 is much faster except eight cases) - [x] JoinBenchmark (JDK11 is faster except five cases) - [x] OrcNestedSchemaPruningBenchmark (JDK11 is faster in nine cases) - [x] PrimitiveArrayBenchmark (JDK11 is faster) - [x] SortBenchmark (JDK11 is faster except `Arrays.sort` case) - [x] UDFBenchmark (N/A, values are too small) - [x] UnsafeArrayDataBenchmark (JDK11 is faster except one case) - [x] WideTableBenchmark (JDK11 is faster except two cases) **B. CASES WE NEED TO INVESTIGATE MORE LATER** - [x] AggregateBenchmark (JDK11 is slower in general) - [x] CompressionSchemeBenchmark (JDK11 is slower in general except `string`) - [x] DataSourceReadBenchmark (JDK11 is slower in general) - [x] DateTimeBenchmark (JDK11 is slightly slower in general except `parsing`) - [x] MakeDateTimeBenchmark (JDK11 is slower except two cases) - [x] MiscBenchmark (JDK11 is slower except ten cases) - [x] OrcV2NestedSchemaPruningBenchmark (JDK11 is slower) - [x] ParquetNestedSchemaPruningBenchmark (JDK11 is slower except six cases) - [x] RangeBenchmark (JDK11 is slower except one case) `FilterPushdownBenchmark/InExpressionBenchmark/WideSchemaBenchmark` will be compared later because it took long timer. ### Why are the changes needed? According to the result, there are some difference between JDK8/JDK11. This will be a baseline for the future improvement and comparison. Also, as a reproducible environment, the following environment is used. - Instance: `r3.xlarge` - OS: `CentOS Linux release 7.5.1804 (Core)` - JDK: - `OpenJDK Runtime Environment (build 1.8.0_222-b10)` - `OpenJDK Runtime Environment 18.9 (build 11.0.4+11-LTS)` ### Does this PR introduce any user-facing change? No. ### How was this patch tested? This is a test-only PR. We need to run benchmark. Closes #26003 from dongjoon-hyun/SPARK-29320. Authored-by: Dongjoon Hyun <dhyun@apple.com> Signed-off-by: Dongjoon Hyun <dhyun@apple.com>
2019-10-03 11:58:25 -04:00
from_utc_timestamp: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative
------------------------------------------------------------------------------------------------------------------------
[SPARK-31630][SQL] Fix perf regression by skipping timestamps rebasing after some threshold ### What changes were proposed in this pull request? Skip timestamps rebasing after a global threshold when there is no difference between Julian and Gregorian calendars. This allows to avoid checking hash maps of switch points, and fixes perf regressions in `toJavaTimestamp()` and `fromJavaTimestamp()`. ### Why are the changes needed? The changes fix perf regressions of conversions to/from external type `java.sql.Timestamp`. Before (see the PR's results https://github.com/apache/spark/pull/28440): ``` ================================================================================================ Conversion from/to external types ================================================================================================ OpenJDK 64-Bit Server VM 1.8.0_252-8u252-b09-1~18.04-b09 on Linux 4.15.0-1063-aws Intel(R) Xeon(R) CPU E5-2670 v2 2.50GHz To/from Java's date-time: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative ------------------------------------------------------------------------------------------------------------------------ From java.sql.Timestamp 376 388 10 13.3 75.2 1.1X Collect java.sql.Timestamp 1878 1937 64 2.7 375.6 0.2X ``` After: ``` ================================================================================================ Conversion from/to external types ================================================================================================ OpenJDK 64-Bit Server VM 1.8.0_252-8u252-b09-1~18.04-b09 on Linux 4.15.0-1063-aws Intel(R) Xeon(R) CPU E5-2670 v2 2.50GHz To/from Java's date-time: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative ------------------------------------------------------------------------------------------------------------------------ From java.sql.Timestamp 249 264 24 20.1 49.8 1.7X Collect java.sql.Timestamp 1503 1523 24 3.3 300.5 0.3X ``` Perf improvements in average of: 1. From java.sql.Timestamp is ~ 34% 2. To java.sql.Timestamps is ~16% ### Does this PR introduce _any_ user-facing change? No ### How was this patch tested? By existing test suites `DateTimeUtilsSuite` and `RebaseDateTimeSuite`. Closes #28441 from MaxGekk/opt-rebase-common-threshold. Authored-by: Max Gekk <max.gekk@gmail.com> Signed-off-by: Wenchen Fan <wenchen@databricks.com>
2020-05-05 10:11:53 -04:00
from_utc_timestamp wholestage off 1316 1320 6 7.6 131.6 1.0X
from_utc_timestamp wholestage on 1268 1272 4 7.9 126.8 1.0X
[SPARK-29320][TESTS] Compare `sql/core` module in JDK8/11 (Part 1) ### What changes were proposed in this pull request? This PR regenerates the `sql/core` benchmarks in JDK8/11 to compare the result. In general, we compare the ratio instead of the time. However, in this PR, the average time is compared. This PR should be considered as a rough comparison. **A. EXPECTED CASES(JDK11 is faster in general)** - [x] BloomFilterBenchmark (JDK11 is faster except one case) - [x] BuiltInDataSourceWriteBenchmark (JDK11 is faster at CSV/ORC) - [x] CSVBenchmark (JDK11 is faster except five cases) - [x] ColumnarBatchBenchmark (JDK11 is faster at `boolean`/`string` and some cases in `int`/`array`) - [x] DatasetBenchmark (JDK11 is faster with `string`, but is slower for `long` type) - [x] ExternalAppendOnlyUnsafeRowArrayBenchmark (JDK11 is faster except two cases) - [x] ExtractBenchmark (JDK11 is faster except HOUR/MINUTE/SECOND/MILLISECONDS/MICROSECONDS) - [x] HashedRelationMetricsBenchmark (JDK11 is faster) - [x] JSONBenchmark (JDK11 is much faster except eight cases) - [x] JoinBenchmark (JDK11 is faster except five cases) - [x] OrcNestedSchemaPruningBenchmark (JDK11 is faster in nine cases) - [x] PrimitiveArrayBenchmark (JDK11 is faster) - [x] SortBenchmark (JDK11 is faster except `Arrays.sort` case) - [x] UDFBenchmark (N/A, values are too small) - [x] UnsafeArrayDataBenchmark (JDK11 is faster except one case) - [x] WideTableBenchmark (JDK11 is faster except two cases) **B. CASES WE NEED TO INVESTIGATE MORE LATER** - [x] AggregateBenchmark (JDK11 is slower in general) - [x] CompressionSchemeBenchmark (JDK11 is slower in general except `string`) - [x] DataSourceReadBenchmark (JDK11 is slower in general) - [x] DateTimeBenchmark (JDK11 is slightly slower in general except `parsing`) - [x] MakeDateTimeBenchmark (JDK11 is slower except two cases) - [x] MiscBenchmark (JDK11 is slower except ten cases) - [x] OrcV2NestedSchemaPruningBenchmark (JDK11 is slower) - [x] ParquetNestedSchemaPruningBenchmark (JDK11 is slower except six cases) - [x] RangeBenchmark (JDK11 is slower except one case) `FilterPushdownBenchmark/InExpressionBenchmark/WideSchemaBenchmark` will be compared later because it took long timer. ### Why are the changes needed? According to the result, there are some difference between JDK8/JDK11. This will be a baseline for the future improvement and comparison. Also, as a reproducible environment, the following environment is used. - Instance: `r3.xlarge` - OS: `CentOS Linux release 7.5.1804 (Core)` - JDK: - `OpenJDK Runtime Environment (build 1.8.0_222-b10)` - `OpenJDK Runtime Environment 18.9 (build 11.0.4+11-LTS)` ### Does this PR introduce any user-facing change? No. ### How was this patch tested? This is a test-only PR. We need to run benchmark. Closes #26003 from dongjoon-hyun/SPARK-29320. Authored-by: Dongjoon Hyun <dhyun@apple.com> Signed-off-by: Dongjoon Hyun <dhyun@apple.com>
2019-10-03 11:58:25 -04:00
[SPARK-31527][SQL][TESTS][FOLLOWUP] Fix the number of rows in `DateTimeBenchmark` ### What changes were proposed in this pull request? - Changed to the number of rows in benchmark cases from 3 to the actual number `N`. - Regenerated benchmark results in the environment: | Item | Description | | ---- | ----| | Region | us-west-2 (Oregon) | | Instance | r3.xlarge | | AMI | ubuntu/images/hvm-ssd/ubuntu-bionic-18.04-amd64-server-20190722.1 (ami-06f2f779464715dc5) | | Java | OpenJDK 64-Bit Server VM 1.8.0_242 and OpenJDK 64-Bit Server VM 11.0.6+10 | ### Why are the changes needed? The changes are needed to have: - Correct benchmark results - Base line for other perf improvements that can be checked in the same environment. ### Does this PR introduce _any_ user-facing change? No ### How was this patch tested? By running the benchmark and checking its output. Closes #28440 from MaxGekk/SPARK-31527-DateTimeBenchmark-followup. Authored-by: Max Gekk <max.gekk@gmail.com> Signed-off-by: Takeshi Yamamuro <yamamuro@apache.org>
2020-05-03 20:39:50 -04:00
OpenJDK 64-Bit Server VM 11.0.7+10-post-Ubuntu-2ubuntu218.04 on Linux 4.15.0-1063-aws
Intel(R) Xeon(R) CPU E5-2670 v2 @ 2.50GHz
[SPARK-29320][TESTS] Compare `sql/core` module in JDK8/11 (Part 1) ### What changes were proposed in this pull request? This PR regenerates the `sql/core` benchmarks in JDK8/11 to compare the result. In general, we compare the ratio instead of the time. However, in this PR, the average time is compared. This PR should be considered as a rough comparison. **A. EXPECTED CASES(JDK11 is faster in general)** - [x] BloomFilterBenchmark (JDK11 is faster except one case) - [x] BuiltInDataSourceWriteBenchmark (JDK11 is faster at CSV/ORC) - [x] CSVBenchmark (JDK11 is faster except five cases) - [x] ColumnarBatchBenchmark (JDK11 is faster at `boolean`/`string` and some cases in `int`/`array`) - [x] DatasetBenchmark (JDK11 is faster with `string`, but is slower for `long` type) - [x] ExternalAppendOnlyUnsafeRowArrayBenchmark (JDK11 is faster except two cases) - [x] ExtractBenchmark (JDK11 is faster except HOUR/MINUTE/SECOND/MILLISECONDS/MICROSECONDS) - [x] HashedRelationMetricsBenchmark (JDK11 is faster) - [x] JSONBenchmark (JDK11 is much faster except eight cases) - [x] JoinBenchmark (JDK11 is faster except five cases) - [x] OrcNestedSchemaPruningBenchmark (JDK11 is faster in nine cases) - [x] PrimitiveArrayBenchmark (JDK11 is faster) - [x] SortBenchmark (JDK11 is faster except `Arrays.sort` case) - [x] UDFBenchmark (N/A, values are too small) - [x] UnsafeArrayDataBenchmark (JDK11 is faster except one case) - [x] WideTableBenchmark (JDK11 is faster except two cases) **B. CASES WE NEED TO INVESTIGATE MORE LATER** - [x] AggregateBenchmark (JDK11 is slower in general) - [x] CompressionSchemeBenchmark (JDK11 is slower in general except `string`) - [x] DataSourceReadBenchmark (JDK11 is slower in general) - [x] DateTimeBenchmark (JDK11 is slightly slower in general except `parsing`) - [x] MakeDateTimeBenchmark (JDK11 is slower except two cases) - [x] MiscBenchmark (JDK11 is slower except ten cases) - [x] OrcV2NestedSchemaPruningBenchmark (JDK11 is slower) - [x] ParquetNestedSchemaPruningBenchmark (JDK11 is slower except six cases) - [x] RangeBenchmark (JDK11 is slower except one case) `FilterPushdownBenchmark/InExpressionBenchmark/WideSchemaBenchmark` will be compared later because it took long timer. ### Why are the changes needed? According to the result, there are some difference between JDK8/JDK11. This will be a baseline for the future improvement and comparison. Also, as a reproducible environment, the following environment is used. - Instance: `r3.xlarge` - OS: `CentOS Linux release 7.5.1804 (Core)` - JDK: - `OpenJDK Runtime Environment (build 1.8.0_222-b10)` - `OpenJDK Runtime Environment 18.9 (build 11.0.4+11-LTS)` ### Does this PR introduce any user-facing change? No. ### How was this patch tested? This is a test-only PR. We need to run benchmark. Closes #26003 from dongjoon-hyun/SPARK-29320. Authored-by: Dongjoon Hyun <dhyun@apple.com> Signed-off-by: Dongjoon Hyun <dhyun@apple.com>
2019-10-03 11:58:25 -04:00
to_utc_timestamp: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative
------------------------------------------------------------------------------------------------------------------------
[SPARK-31630][SQL] Fix perf regression by skipping timestamps rebasing after some threshold ### What changes were proposed in this pull request? Skip timestamps rebasing after a global threshold when there is no difference between Julian and Gregorian calendars. This allows to avoid checking hash maps of switch points, and fixes perf regressions in `toJavaTimestamp()` and `fromJavaTimestamp()`. ### Why are the changes needed? The changes fix perf regressions of conversions to/from external type `java.sql.Timestamp`. Before (see the PR's results https://github.com/apache/spark/pull/28440): ``` ================================================================================================ Conversion from/to external types ================================================================================================ OpenJDK 64-Bit Server VM 1.8.0_252-8u252-b09-1~18.04-b09 on Linux 4.15.0-1063-aws Intel(R) Xeon(R) CPU E5-2670 v2 2.50GHz To/from Java's date-time: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative ------------------------------------------------------------------------------------------------------------------------ From java.sql.Timestamp 376 388 10 13.3 75.2 1.1X Collect java.sql.Timestamp 1878 1937 64 2.7 375.6 0.2X ``` After: ``` ================================================================================================ Conversion from/to external types ================================================================================================ OpenJDK 64-Bit Server VM 1.8.0_252-8u252-b09-1~18.04-b09 on Linux 4.15.0-1063-aws Intel(R) Xeon(R) CPU E5-2670 v2 2.50GHz To/from Java's date-time: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative ------------------------------------------------------------------------------------------------------------------------ From java.sql.Timestamp 249 264 24 20.1 49.8 1.7X Collect java.sql.Timestamp 1503 1523 24 3.3 300.5 0.3X ``` Perf improvements in average of: 1. From java.sql.Timestamp is ~ 34% 2. To java.sql.Timestamps is ~16% ### Does this PR introduce _any_ user-facing change? No ### How was this patch tested? By existing test suites `DateTimeUtilsSuite` and `RebaseDateTimeSuite`. Closes #28441 from MaxGekk/opt-rebase-common-threshold. Authored-by: Max Gekk <max.gekk@gmail.com> Signed-off-by: Wenchen Fan <wenchen@databricks.com>
2020-05-05 10:11:53 -04:00
to_utc_timestamp wholestage off 1653 1657 6 6.0 165.3 1.0X
to_utc_timestamp wholestage on 1594 1599 4 6.3 159.4 1.0X
[SPARK-29320][TESTS] Compare `sql/core` module in JDK8/11 (Part 1) ### What changes were proposed in this pull request? This PR regenerates the `sql/core` benchmarks in JDK8/11 to compare the result. In general, we compare the ratio instead of the time. However, in this PR, the average time is compared. This PR should be considered as a rough comparison. **A. EXPECTED CASES(JDK11 is faster in general)** - [x] BloomFilterBenchmark (JDK11 is faster except one case) - [x] BuiltInDataSourceWriteBenchmark (JDK11 is faster at CSV/ORC) - [x] CSVBenchmark (JDK11 is faster except five cases) - [x] ColumnarBatchBenchmark (JDK11 is faster at `boolean`/`string` and some cases in `int`/`array`) - [x] DatasetBenchmark (JDK11 is faster with `string`, but is slower for `long` type) - [x] ExternalAppendOnlyUnsafeRowArrayBenchmark (JDK11 is faster except two cases) - [x] ExtractBenchmark (JDK11 is faster except HOUR/MINUTE/SECOND/MILLISECONDS/MICROSECONDS) - [x] HashedRelationMetricsBenchmark (JDK11 is faster) - [x] JSONBenchmark (JDK11 is much faster except eight cases) - [x] JoinBenchmark (JDK11 is faster except five cases) - [x] OrcNestedSchemaPruningBenchmark (JDK11 is faster in nine cases) - [x] PrimitiveArrayBenchmark (JDK11 is faster) - [x] SortBenchmark (JDK11 is faster except `Arrays.sort` case) - [x] UDFBenchmark (N/A, values are too small) - [x] UnsafeArrayDataBenchmark (JDK11 is faster except one case) - [x] WideTableBenchmark (JDK11 is faster except two cases) **B. CASES WE NEED TO INVESTIGATE MORE LATER** - [x] AggregateBenchmark (JDK11 is slower in general) - [x] CompressionSchemeBenchmark (JDK11 is slower in general except `string`) - [x] DataSourceReadBenchmark (JDK11 is slower in general) - [x] DateTimeBenchmark (JDK11 is slightly slower in general except `parsing`) - [x] MakeDateTimeBenchmark (JDK11 is slower except two cases) - [x] MiscBenchmark (JDK11 is slower except ten cases) - [x] OrcV2NestedSchemaPruningBenchmark (JDK11 is slower) - [x] ParquetNestedSchemaPruningBenchmark (JDK11 is slower except six cases) - [x] RangeBenchmark (JDK11 is slower except one case) `FilterPushdownBenchmark/InExpressionBenchmark/WideSchemaBenchmark` will be compared later because it took long timer. ### Why are the changes needed? According to the result, there are some difference between JDK8/JDK11. This will be a baseline for the future improvement and comparison. Also, as a reproducible environment, the following environment is used. - Instance: `r3.xlarge` - OS: `CentOS Linux release 7.5.1804 (Core)` - JDK: - `OpenJDK Runtime Environment (build 1.8.0_222-b10)` - `OpenJDK Runtime Environment 18.9 (build 11.0.4+11-LTS)` ### Does this PR introduce any user-facing change? No. ### How was this patch tested? This is a test-only PR. We need to run benchmark. Closes #26003 from dongjoon-hyun/SPARK-29320. Authored-by: Dongjoon Hyun <dhyun@apple.com> Signed-off-by: Dongjoon Hyun <dhyun@apple.com>
2019-10-03 11:58:25 -04:00
================================================================================================
Intervals
================================================================================================
[SPARK-31527][SQL][TESTS][FOLLOWUP] Fix the number of rows in `DateTimeBenchmark` ### What changes were proposed in this pull request? - Changed to the number of rows in benchmark cases from 3 to the actual number `N`. - Regenerated benchmark results in the environment: | Item | Description | | ---- | ----| | Region | us-west-2 (Oregon) | | Instance | r3.xlarge | | AMI | ubuntu/images/hvm-ssd/ubuntu-bionic-18.04-amd64-server-20190722.1 (ami-06f2f779464715dc5) | | Java | OpenJDK 64-Bit Server VM 1.8.0_242 and OpenJDK 64-Bit Server VM 11.0.6+10 | ### Why are the changes needed? The changes are needed to have: - Correct benchmark results - Base line for other perf improvements that can be checked in the same environment. ### Does this PR introduce _any_ user-facing change? No ### How was this patch tested? By running the benchmark and checking its output. Closes #28440 from MaxGekk/SPARK-31527-DateTimeBenchmark-followup. Authored-by: Max Gekk <max.gekk@gmail.com> Signed-off-by: Takeshi Yamamuro <yamamuro@apache.org>
2020-05-03 20:39:50 -04:00
OpenJDK 64-Bit Server VM 11.0.7+10-post-Ubuntu-2ubuntu218.04 on Linux 4.15.0-1063-aws
Intel(R) Xeon(R) CPU E5-2670 v2 @ 2.50GHz
[SPARK-29320][TESTS] Compare `sql/core` module in JDK8/11 (Part 1) ### What changes were proposed in this pull request? This PR regenerates the `sql/core` benchmarks in JDK8/11 to compare the result. In general, we compare the ratio instead of the time. However, in this PR, the average time is compared. This PR should be considered as a rough comparison. **A. EXPECTED CASES(JDK11 is faster in general)** - [x] BloomFilterBenchmark (JDK11 is faster except one case) - [x] BuiltInDataSourceWriteBenchmark (JDK11 is faster at CSV/ORC) - [x] CSVBenchmark (JDK11 is faster except five cases) - [x] ColumnarBatchBenchmark (JDK11 is faster at `boolean`/`string` and some cases in `int`/`array`) - [x] DatasetBenchmark (JDK11 is faster with `string`, but is slower for `long` type) - [x] ExternalAppendOnlyUnsafeRowArrayBenchmark (JDK11 is faster except two cases) - [x] ExtractBenchmark (JDK11 is faster except HOUR/MINUTE/SECOND/MILLISECONDS/MICROSECONDS) - [x] HashedRelationMetricsBenchmark (JDK11 is faster) - [x] JSONBenchmark (JDK11 is much faster except eight cases) - [x] JoinBenchmark (JDK11 is faster except five cases) - [x] OrcNestedSchemaPruningBenchmark (JDK11 is faster in nine cases) - [x] PrimitiveArrayBenchmark (JDK11 is faster) - [x] SortBenchmark (JDK11 is faster except `Arrays.sort` case) - [x] UDFBenchmark (N/A, values are too small) - [x] UnsafeArrayDataBenchmark (JDK11 is faster except one case) - [x] WideTableBenchmark (JDK11 is faster except two cases) **B. CASES WE NEED TO INVESTIGATE MORE LATER** - [x] AggregateBenchmark (JDK11 is slower in general) - [x] CompressionSchemeBenchmark (JDK11 is slower in general except `string`) - [x] DataSourceReadBenchmark (JDK11 is slower in general) - [x] DateTimeBenchmark (JDK11 is slightly slower in general except `parsing`) - [x] MakeDateTimeBenchmark (JDK11 is slower except two cases) - [x] MiscBenchmark (JDK11 is slower except ten cases) - [x] OrcV2NestedSchemaPruningBenchmark (JDK11 is slower) - [x] ParquetNestedSchemaPruningBenchmark (JDK11 is slower except six cases) - [x] RangeBenchmark (JDK11 is slower except one case) `FilterPushdownBenchmark/InExpressionBenchmark/WideSchemaBenchmark` will be compared later because it took long timer. ### Why are the changes needed? According to the result, there are some difference between JDK8/JDK11. This will be a baseline for the future improvement and comparison. Also, as a reproducible environment, the following environment is used. - Instance: `r3.xlarge` - OS: `CentOS Linux release 7.5.1804 (Core)` - JDK: - `OpenJDK Runtime Environment (build 1.8.0_222-b10)` - `OpenJDK Runtime Environment 18.9 (build 11.0.4+11-LTS)` ### Does this PR introduce any user-facing change? No. ### How was this patch tested? This is a test-only PR. We need to run benchmark. Closes #26003 from dongjoon-hyun/SPARK-29320. Authored-by: Dongjoon Hyun <dhyun@apple.com> Signed-off-by: Dongjoon Hyun <dhyun@apple.com>
2019-10-03 11:58:25 -04:00
cast interval: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative
------------------------------------------------------------------------------------------------------------------------
[SPARK-31630][SQL] Fix perf regression by skipping timestamps rebasing after some threshold ### What changes were proposed in this pull request? Skip timestamps rebasing after a global threshold when there is no difference between Julian and Gregorian calendars. This allows to avoid checking hash maps of switch points, and fixes perf regressions in `toJavaTimestamp()` and `fromJavaTimestamp()`. ### Why are the changes needed? The changes fix perf regressions of conversions to/from external type `java.sql.Timestamp`. Before (see the PR's results https://github.com/apache/spark/pull/28440): ``` ================================================================================================ Conversion from/to external types ================================================================================================ OpenJDK 64-Bit Server VM 1.8.0_252-8u252-b09-1~18.04-b09 on Linux 4.15.0-1063-aws Intel(R) Xeon(R) CPU E5-2670 v2 2.50GHz To/from Java's date-time: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative ------------------------------------------------------------------------------------------------------------------------ From java.sql.Timestamp 376 388 10 13.3 75.2 1.1X Collect java.sql.Timestamp 1878 1937 64 2.7 375.6 0.2X ``` After: ``` ================================================================================================ Conversion from/to external types ================================================================================================ OpenJDK 64-Bit Server VM 1.8.0_252-8u252-b09-1~18.04-b09 on Linux 4.15.0-1063-aws Intel(R) Xeon(R) CPU E5-2670 v2 2.50GHz To/from Java's date-time: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative ------------------------------------------------------------------------------------------------------------------------ From java.sql.Timestamp 249 264 24 20.1 49.8 1.7X Collect java.sql.Timestamp 1503 1523 24 3.3 300.5 0.3X ``` Perf improvements in average of: 1. From java.sql.Timestamp is ~ 34% 2. To java.sql.Timestamps is ~16% ### Does this PR introduce _any_ user-facing change? No ### How was this patch tested? By existing test suites `DateTimeUtilsSuite` and `RebaseDateTimeSuite`. Closes #28441 from MaxGekk/opt-rebase-common-threshold. Authored-by: Max Gekk <max.gekk@gmail.com> Signed-off-by: Wenchen Fan <wenchen@databricks.com>
2020-05-05 10:11:53 -04:00
cast interval wholestage off 341 343 3 29.4 34.1 1.0X
cast interval wholestage on 279 282 1 35.8 27.9 1.2X
[SPARK-29320][TESTS] Compare `sql/core` module in JDK8/11 (Part 1) ### What changes were proposed in this pull request? This PR regenerates the `sql/core` benchmarks in JDK8/11 to compare the result. In general, we compare the ratio instead of the time. However, in this PR, the average time is compared. This PR should be considered as a rough comparison. **A. EXPECTED CASES(JDK11 is faster in general)** - [x] BloomFilterBenchmark (JDK11 is faster except one case) - [x] BuiltInDataSourceWriteBenchmark (JDK11 is faster at CSV/ORC) - [x] CSVBenchmark (JDK11 is faster except five cases) - [x] ColumnarBatchBenchmark (JDK11 is faster at `boolean`/`string` and some cases in `int`/`array`) - [x] DatasetBenchmark (JDK11 is faster with `string`, but is slower for `long` type) - [x] ExternalAppendOnlyUnsafeRowArrayBenchmark (JDK11 is faster except two cases) - [x] ExtractBenchmark (JDK11 is faster except HOUR/MINUTE/SECOND/MILLISECONDS/MICROSECONDS) - [x] HashedRelationMetricsBenchmark (JDK11 is faster) - [x] JSONBenchmark (JDK11 is much faster except eight cases) - [x] JoinBenchmark (JDK11 is faster except five cases) - [x] OrcNestedSchemaPruningBenchmark (JDK11 is faster in nine cases) - [x] PrimitiveArrayBenchmark (JDK11 is faster) - [x] SortBenchmark (JDK11 is faster except `Arrays.sort` case) - [x] UDFBenchmark (N/A, values are too small) - [x] UnsafeArrayDataBenchmark (JDK11 is faster except one case) - [x] WideTableBenchmark (JDK11 is faster except two cases) **B. CASES WE NEED TO INVESTIGATE MORE LATER** - [x] AggregateBenchmark (JDK11 is slower in general) - [x] CompressionSchemeBenchmark (JDK11 is slower in general except `string`) - [x] DataSourceReadBenchmark (JDK11 is slower in general) - [x] DateTimeBenchmark (JDK11 is slightly slower in general except `parsing`) - [x] MakeDateTimeBenchmark (JDK11 is slower except two cases) - [x] MiscBenchmark (JDK11 is slower except ten cases) - [x] OrcV2NestedSchemaPruningBenchmark (JDK11 is slower) - [x] ParquetNestedSchemaPruningBenchmark (JDK11 is slower except six cases) - [x] RangeBenchmark (JDK11 is slower except one case) `FilterPushdownBenchmark/InExpressionBenchmark/WideSchemaBenchmark` will be compared later because it took long timer. ### Why are the changes needed? According to the result, there are some difference between JDK8/JDK11. This will be a baseline for the future improvement and comparison. Also, as a reproducible environment, the following environment is used. - Instance: `r3.xlarge` - OS: `CentOS Linux release 7.5.1804 (Core)` - JDK: - `OpenJDK Runtime Environment (build 1.8.0_222-b10)` - `OpenJDK Runtime Environment 18.9 (build 11.0.4+11-LTS)` ### Does this PR introduce any user-facing change? No. ### How was this patch tested? This is a test-only PR. We need to run benchmark. Closes #26003 from dongjoon-hyun/SPARK-29320. Authored-by: Dongjoon Hyun <dhyun@apple.com> Signed-off-by: Dongjoon Hyun <dhyun@apple.com>
2019-10-03 11:58:25 -04:00
[SPARK-31527][SQL][TESTS][FOLLOWUP] Fix the number of rows in `DateTimeBenchmark` ### What changes were proposed in this pull request? - Changed to the number of rows in benchmark cases from 3 to the actual number `N`. - Regenerated benchmark results in the environment: | Item | Description | | ---- | ----| | Region | us-west-2 (Oregon) | | Instance | r3.xlarge | | AMI | ubuntu/images/hvm-ssd/ubuntu-bionic-18.04-amd64-server-20190722.1 (ami-06f2f779464715dc5) | | Java | OpenJDK 64-Bit Server VM 1.8.0_242 and OpenJDK 64-Bit Server VM 11.0.6+10 | ### Why are the changes needed? The changes are needed to have: - Correct benchmark results - Base line for other perf improvements that can be checked in the same environment. ### Does this PR introduce _any_ user-facing change? No ### How was this patch tested? By running the benchmark and checking its output. Closes #28440 from MaxGekk/SPARK-31527-DateTimeBenchmark-followup. Authored-by: Max Gekk <max.gekk@gmail.com> Signed-off-by: Takeshi Yamamuro <yamamuro@apache.org>
2020-05-03 20:39:50 -04:00
OpenJDK 64-Bit Server VM 11.0.7+10-post-Ubuntu-2ubuntu218.04 on Linux 4.15.0-1063-aws
Intel(R) Xeon(R) CPU E5-2670 v2 @ 2.50GHz
[SPARK-29320][TESTS] Compare `sql/core` module in JDK8/11 (Part 1) ### What changes were proposed in this pull request? This PR regenerates the `sql/core` benchmarks in JDK8/11 to compare the result. In general, we compare the ratio instead of the time. However, in this PR, the average time is compared. This PR should be considered as a rough comparison. **A. EXPECTED CASES(JDK11 is faster in general)** - [x] BloomFilterBenchmark (JDK11 is faster except one case) - [x] BuiltInDataSourceWriteBenchmark (JDK11 is faster at CSV/ORC) - [x] CSVBenchmark (JDK11 is faster except five cases) - [x] ColumnarBatchBenchmark (JDK11 is faster at `boolean`/`string` and some cases in `int`/`array`) - [x] DatasetBenchmark (JDK11 is faster with `string`, but is slower for `long` type) - [x] ExternalAppendOnlyUnsafeRowArrayBenchmark (JDK11 is faster except two cases) - [x] ExtractBenchmark (JDK11 is faster except HOUR/MINUTE/SECOND/MILLISECONDS/MICROSECONDS) - [x] HashedRelationMetricsBenchmark (JDK11 is faster) - [x] JSONBenchmark (JDK11 is much faster except eight cases) - [x] JoinBenchmark (JDK11 is faster except five cases) - [x] OrcNestedSchemaPruningBenchmark (JDK11 is faster in nine cases) - [x] PrimitiveArrayBenchmark (JDK11 is faster) - [x] SortBenchmark (JDK11 is faster except `Arrays.sort` case) - [x] UDFBenchmark (N/A, values are too small) - [x] UnsafeArrayDataBenchmark (JDK11 is faster except one case) - [x] WideTableBenchmark (JDK11 is faster except two cases) **B. CASES WE NEED TO INVESTIGATE MORE LATER** - [x] AggregateBenchmark (JDK11 is slower in general) - [x] CompressionSchemeBenchmark (JDK11 is slower in general except `string`) - [x] DataSourceReadBenchmark (JDK11 is slower in general) - [x] DateTimeBenchmark (JDK11 is slightly slower in general except `parsing`) - [x] MakeDateTimeBenchmark (JDK11 is slower except two cases) - [x] MiscBenchmark (JDK11 is slower except ten cases) - [x] OrcV2NestedSchemaPruningBenchmark (JDK11 is slower) - [x] ParquetNestedSchemaPruningBenchmark (JDK11 is slower except six cases) - [x] RangeBenchmark (JDK11 is slower except one case) `FilterPushdownBenchmark/InExpressionBenchmark/WideSchemaBenchmark` will be compared later because it took long timer. ### Why are the changes needed? According to the result, there are some difference between JDK8/JDK11. This will be a baseline for the future improvement and comparison. Also, as a reproducible environment, the following environment is used. - Instance: `r3.xlarge` - OS: `CentOS Linux release 7.5.1804 (Core)` - JDK: - `OpenJDK Runtime Environment (build 1.8.0_222-b10)` - `OpenJDK Runtime Environment 18.9 (build 11.0.4+11-LTS)` ### Does this PR introduce any user-facing change? No. ### How was this patch tested? This is a test-only PR. We need to run benchmark. Closes #26003 from dongjoon-hyun/SPARK-29320. Authored-by: Dongjoon Hyun <dhyun@apple.com> Signed-off-by: Dongjoon Hyun <dhyun@apple.com>
2019-10-03 11:58:25 -04:00
datediff: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative
------------------------------------------------------------------------------------------------------------------------
[SPARK-31630][SQL] Fix perf regression by skipping timestamps rebasing after some threshold ### What changes were proposed in this pull request? Skip timestamps rebasing after a global threshold when there is no difference between Julian and Gregorian calendars. This allows to avoid checking hash maps of switch points, and fixes perf regressions in `toJavaTimestamp()` and `fromJavaTimestamp()`. ### Why are the changes needed? The changes fix perf regressions of conversions to/from external type `java.sql.Timestamp`. Before (see the PR's results https://github.com/apache/spark/pull/28440): ``` ================================================================================================ Conversion from/to external types ================================================================================================ OpenJDK 64-Bit Server VM 1.8.0_252-8u252-b09-1~18.04-b09 on Linux 4.15.0-1063-aws Intel(R) Xeon(R) CPU E5-2670 v2 2.50GHz To/from Java's date-time: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative ------------------------------------------------------------------------------------------------------------------------ From java.sql.Timestamp 376 388 10 13.3 75.2 1.1X Collect java.sql.Timestamp 1878 1937 64 2.7 375.6 0.2X ``` After: ``` ================================================================================================ Conversion from/to external types ================================================================================================ OpenJDK 64-Bit Server VM 1.8.0_252-8u252-b09-1~18.04-b09 on Linux 4.15.0-1063-aws Intel(R) Xeon(R) CPU E5-2670 v2 2.50GHz To/from Java's date-time: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative ------------------------------------------------------------------------------------------------------------------------ From java.sql.Timestamp 249 264 24 20.1 49.8 1.7X Collect java.sql.Timestamp 1503 1523 24 3.3 300.5 0.3X ``` Perf improvements in average of: 1. From java.sql.Timestamp is ~ 34% 2. To java.sql.Timestamps is ~16% ### Does this PR introduce _any_ user-facing change? No ### How was this patch tested? By existing test suites `DateTimeUtilsSuite` and `RebaseDateTimeSuite`. Closes #28441 from MaxGekk/opt-rebase-common-threshold. Authored-by: Max Gekk <max.gekk@gmail.com> Signed-off-by: Wenchen Fan <wenchen@databricks.com>
2020-05-05 10:11:53 -04:00
datediff wholestage off 1862 1865 4 5.4 186.2 1.0X
datediff wholestage on 1769 1783 15 5.7 176.9 1.1X
[SPARK-29320][TESTS] Compare `sql/core` module in JDK8/11 (Part 1) ### What changes were proposed in this pull request? This PR regenerates the `sql/core` benchmarks in JDK8/11 to compare the result. In general, we compare the ratio instead of the time. However, in this PR, the average time is compared. This PR should be considered as a rough comparison. **A. EXPECTED CASES(JDK11 is faster in general)** - [x] BloomFilterBenchmark (JDK11 is faster except one case) - [x] BuiltInDataSourceWriteBenchmark (JDK11 is faster at CSV/ORC) - [x] CSVBenchmark (JDK11 is faster except five cases) - [x] ColumnarBatchBenchmark (JDK11 is faster at `boolean`/`string` and some cases in `int`/`array`) - [x] DatasetBenchmark (JDK11 is faster with `string`, but is slower for `long` type) - [x] ExternalAppendOnlyUnsafeRowArrayBenchmark (JDK11 is faster except two cases) - [x] ExtractBenchmark (JDK11 is faster except HOUR/MINUTE/SECOND/MILLISECONDS/MICROSECONDS) - [x] HashedRelationMetricsBenchmark (JDK11 is faster) - [x] JSONBenchmark (JDK11 is much faster except eight cases) - [x] JoinBenchmark (JDK11 is faster except five cases) - [x] OrcNestedSchemaPruningBenchmark (JDK11 is faster in nine cases) - [x] PrimitiveArrayBenchmark (JDK11 is faster) - [x] SortBenchmark (JDK11 is faster except `Arrays.sort` case) - [x] UDFBenchmark (N/A, values are too small) - [x] UnsafeArrayDataBenchmark (JDK11 is faster except one case) - [x] WideTableBenchmark (JDK11 is faster except two cases) **B. CASES WE NEED TO INVESTIGATE MORE LATER** - [x] AggregateBenchmark (JDK11 is slower in general) - [x] CompressionSchemeBenchmark (JDK11 is slower in general except `string`) - [x] DataSourceReadBenchmark (JDK11 is slower in general) - [x] DateTimeBenchmark (JDK11 is slightly slower in general except `parsing`) - [x] MakeDateTimeBenchmark (JDK11 is slower except two cases) - [x] MiscBenchmark (JDK11 is slower except ten cases) - [x] OrcV2NestedSchemaPruningBenchmark (JDK11 is slower) - [x] ParquetNestedSchemaPruningBenchmark (JDK11 is slower except six cases) - [x] RangeBenchmark (JDK11 is slower except one case) `FilterPushdownBenchmark/InExpressionBenchmark/WideSchemaBenchmark` will be compared later because it took long timer. ### Why are the changes needed? According to the result, there are some difference between JDK8/JDK11. This will be a baseline for the future improvement and comparison. Also, as a reproducible environment, the following environment is used. - Instance: `r3.xlarge` - OS: `CentOS Linux release 7.5.1804 (Core)` - JDK: - `OpenJDK Runtime Environment (build 1.8.0_222-b10)` - `OpenJDK Runtime Environment 18.9 (build 11.0.4+11-LTS)` ### Does this PR introduce any user-facing change? No. ### How was this patch tested? This is a test-only PR. We need to run benchmark. Closes #26003 from dongjoon-hyun/SPARK-29320. Authored-by: Dongjoon Hyun <dhyun@apple.com> Signed-off-by: Dongjoon Hyun <dhyun@apple.com>
2019-10-03 11:58:25 -04:00
[SPARK-31527][SQL][TESTS][FOLLOWUP] Fix the number of rows in `DateTimeBenchmark` ### What changes were proposed in this pull request? - Changed to the number of rows in benchmark cases from 3 to the actual number `N`. - Regenerated benchmark results in the environment: | Item | Description | | ---- | ----| | Region | us-west-2 (Oregon) | | Instance | r3.xlarge | | AMI | ubuntu/images/hvm-ssd/ubuntu-bionic-18.04-amd64-server-20190722.1 (ami-06f2f779464715dc5) | | Java | OpenJDK 64-Bit Server VM 1.8.0_242 and OpenJDK 64-Bit Server VM 11.0.6+10 | ### Why are the changes needed? The changes are needed to have: - Correct benchmark results - Base line for other perf improvements that can be checked in the same environment. ### Does this PR introduce _any_ user-facing change? No ### How was this patch tested? By running the benchmark and checking its output. Closes #28440 from MaxGekk/SPARK-31527-DateTimeBenchmark-followup. Authored-by: Max Gekk <max.gekk@gmail.com> Signed-off-by: Takeshi Yamamuro <yamamuro@apache.org>
2020-05-03 20:39:50 -04:00
OpenJDK 64-Bit Server VM 11.0.7+10-post-Ubuntu-2ubuntu218.04 on Linux 4.15.0-1063-aws
Intel(R) Xeon(R) CPU E5-2670 v2 @ 2.50GHz
[SPARK-29320][TESTS] Compare `sql/core` module in JDK8/11 (Part 1) ### What changes were proposed in this pull request? This PR regenerates the `sql/core` benchmarks in JDK8/11 to compare the result. In general, we compare the ratio instead of the time. However, in this PR, the average time is compared. This PR should be considered as a rough comparison. **A. EXPECTED CASES(JDK11 is faster in general)** - [x] BloomFilterBenchmark (JDK11 is faster except one case) - [x] BuiltInDataSourceWriteBenchmark (JDK11 is faster at CSV/ORC) - [x] CSVBenchmark (JDK11 is faster except five cases) - [x] ColumnarBatchBenchmark (JDK11 is faster at `boolean`/`string` and some cases in `int`/`array`) - [x] DatasetBenchmark (JDK11 is faster with `string`, but is slower for `long` type) - [x] ExternalAppendOnlyUnsafeRowArrayBenchmark (JDK11 is faster except two cases) - [x] ExtractBenchmark (JDK11 is faster except HOUR/MINUTE/SECOND/MILLISECONDS/MICROSECONDS) - [x] HashedRelationMetricsBenchmark (JDK11 is faster) - [x] JSONBenchmark (JDK11 is much faster except eight cases) - [x] JoinBenchmark (JDK11 is faster except five cases) - [x] OrcNestedSchemaPruningBenchmark (JDK11 is faster in nine cases) - [x] PrimitiveArrayBenchmark (JDK11 is faster) - [x] SortBenchmark (JDK11 is faster except `Arrays.sort` case) - [x] UDFBenchmark (N/A, values are too small) - [x] UnsafeArrayDataBenchmark (JDK11 is faster except one case) - [x] WideTableBenchmark (JDK11 is faster except two cases) **B. CASES WE NEED TO INVESTIGATE MORE LATER** - [x] AggregateBenchmark (JDK11 is slower in general) - [x] CompressionSchemeBenchmark (JDK11 is slower in general except `string`) - [x] DataSourceReadBenchmark (JDK11 is slower in general) - [x] DateTimeBenchmark (JDK11 is slightly slower in general except `parsing`) - [x] MakeDateTimeBenchmark (JDK11 is slower except two cases) - [x] MiscBenchmark (JDK11 is slower except ten cases) - [x] OrcV2NestedSchemaPruningBenchmark (JDK11 is slower) - [x] ParquetNestedSchemaPruningBenchmark (JDK11 is slower except six cases) - [x] RangeBenchmark (JDK11 is slower except one case) `FilterPushdownBenchmark/InExpressionBenchmark/WideSchemaBenchmark` will be compared later because it took long timer. ### Why are the changes needed? According to the result, there are some difference between JDK8/JDK11. This will be a baseline for the future improvement and comparison. Also, as a reproducible environment, the following environment is used. - Instance: `r3.xlarge` - OS: `CentOS Linux release 7.5.1804 (Core)` - JDK: - `OpenJDK Runtime Environment (build 1.8.0_222-b10)` - `OpenJDK Runtime Environment 18.9 (build 11.0.4+11-LTS)` ### Does this PR introduce any user-facing change? No. ### How was this patch tested? This is a test-only PR. We need to run benchmark. Closes #26003 from dongjoon-hyun/SPARK-29320. Authored-by: Dongjoon Hyun <dhyun@apple.com> Signed-off-by: Dongjoon Hyun <dhyun@apple.com>
2019-10-03 11:58:25 -04:00
months_between: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative
------------------------------------------------------------------------------------------------------------------------
[SPARK-31630][SQL] Fix perf regression by skipping timestamps rebasing after some threshold ### What changes were proposed in this pull request? Skip timestamps rebasing after a global threshold when there is no difference between Julian and Gregorian calendars. This allows to avoid checking hash maps of switch points, and fixes perf regressions in `toJavaTimestamp()` and `fromJavaTimestamp()`. ### Why are the changes needed? The changes fix perf regressions of conversions to/from external type `java.sql.Timestamp`. Before (see the PR's results https://github.com/apache/spark/pull/28440): ``` ================================================================================================ Conversion from/to external types ================================================================================================ OpenJDK 64-Bit Server VM 1.8.0_252-8u252-b09-1~18.04-b09 on Linux 4.15.0-1063-aws Intel(R) Xeon(R) CPU E5-2670 v2 2.50GHz To/from Java's date-time: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative ------------------------------------------------------------------------------------------------------------------------ From java.sql.Timestamp 376 388 10 13.3 75.2 1.1X Collect java.sql.Timestamp 1878 1937 64 2.7 375.6 0.2X ``` After: ``` ================================================================================================ Conversion from/to external types ================================================================================================ OpenJDK 64-Bit Server VM 1.8.0_252-8u252-b09-1~18.04-b09 on Linux 4.15.0-1063-aws Intel(R) Xeon(R) CPU E5-2670 v2 2.50GHz To/from Java's date-time: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative ------------------------------------------------------------------------------------------------------------------------ From java.sql.Timestamp 249 264 24 20.1 49.8 1.7X Collect java.sql.Timestamp 1503 1523 24 3.3 300.5 0.3X ``` Perf improvements in average of: 1. From java.sql.Timestamp is ~ 34% 2. To java.sql.Timestamps is ~16% ### Does this PR introduce _any_ user-facing change? No ### How was this patch tested? By existing test suites `DateTimeUtilsSuite` and `RebaseDateTimeSuite`. Closes #28441 from MaxGekk/opt-rebase-common-threshold. Authored-by: Max Gekk <max.gekk@gmail.com> Signed-off-by: Wenchen Fan <wenchen@databricks.com>
2020-05-05 10:11:53 -04:00
months_between wholestage off 5594 5599 7 1.8 559.4 1.0X
months_between wholestage on 5498 5508 11 1.8 549.8 1.0X
[SPARK-29320][TESTS] Compare `sql/core` module in JDK8/11 (Part 1) ### What changes were proposed in this pull request? This PR regenerates the `sql/core` benchmarks in JDK8/11 to compare the result. In general, we compare the ratio instead of the time. However, in this PR, the average time is compared. This PR should be considered as a rough comparison. **A. EXPECTED CASES(JDK11 is faster in general)** - [x] BloomFilterBenchmark (JDK11 is faster except one case) - [x] BuiltInDataSourceWriteBenchmark (JDK11 is faster at CSV/ORC) - [x] CSVBenchmark (JDK11 is faster except five cases) - [x] ColumnarBatchBenchmark (JDK11 is faster at `boolean`/`string` and some cases in `int`/`array`) - [x] DatasetBenchmark (JDK11 is faster with `string`, but is slower for `long` type) - [x] ExternalAppendOnlyUnsafeRowArrayBenchmark (JDK11 is faster except two cases) - [x] ExtractBenchmark (JDK11 is faster except HOUR/MINUTE/SECOND/MILLISECONDS/MICROSECONDS) - [x] HashedRelationMetricsBenchmark (JDK11 is faster) - [x] JSONBenchmark (JDK11 is much faster except eight cases) - [x] JoinBenchmark (JDK11 is faster except five cases) - [x] OrcNestedSchemaPruningBenchmark (JDK11 is faster in nine cases) - [x] PrimitiveArrayBenchmark (JDK11 is faster) - [x] SortBenchmark (JDK11 is faster except `Arrays.sort` case) - [x] UDFBenchmark (N/A, values are too small) - [x] UnsafeArrayDataBenchmark (JDK11 is faster except one case) - [x] WideTableBenchmark (JDK11 is faster except two cases) **B. CASES WE NEED TO INVESTIGATE MORE LATER** - [x] AggregateBenchmark (JDK11 is slower in general) - [x] CompressionSchemeBenchmark (JDK11 is slower in general except `string`) - [x] DataSourceReadBenchmark (JDK11 is slower in general) - [x] DateTimeBenchmark (JDK11 is slightly slower in general except `parsing`) - [x] MakeDateTimeBenchmark (JDK11 is slower except two cases) - [x] MiscBenchmark (JDK11 is slower except ten cases) - [x] OrcV2NestedSchemaPruningBenchmark (JDK11 is slower) - [x] ParquetNestedSchemaPruningBenchmark (JDK11 is slower except six cases) - [x] RangeBenchmark (JDK11 is slower except one case) `FilterPushdownBenchmark/InExpressionBenchmark/WideSchemaBenchmark` will be compared later because it took long timer. ### Why are the changes needed? According to the result, there are some difference between JDK8/JDK11. This will be a baseline for the future improvement and comparison. Also, as a reproducible environment, the following environment is used. - Instance: `r3.xlarge` - OS: `CentOS Linux release 7.5.1804 (Core)` - JDK: - `OpenJDK Runtime Environment (build 1.8.0_222-b10)` - `OpenJDK Runtime Environment 18.9 (build 11.0.4+11-LTS)` ### Does this PR introduce any user-facing change? No. ### How was this patch tested? This is a test-only PR. We need to run benchmark. Closes #26003 from dongjoon-hyun/SPARK-29320. Authored-by: Dongjoon Hyun <dhyun@apple.com> Signed-off-by: Dongjoon Hyun <dhyun@apple.com>
2019-10-03 11:58:25 -04:00
[SPARK-31527][SQL][TESTS][FOLLOWUP] Fix the number of rows in `DateTimeBenchmark` ### What changes were proposed in this pull request? - Changed to the number of rows in benchmark cases from 3 to the actual number `N`. - Regenerated benchmark results in the environment: | Item | Description | | ---- | ----| | Region | us-west-2 (Oregon) | | Instance | r3.xlarge | | AMI | ubuntu/images/hvm-ssd/ubuntu-bionic-18.04-amd64-server-20190722.1 (ami-06f2f779464715dc5) | | Java | OpenJDK 64-Bit Server VM 1.8.0_242 and OpenJDK 64-Bit Server VM 11.0.6+10 | ### Why are the changes needed? The changes are needed to have: - Correct benchmark results - Base line for other perf improvements that can be checked in the same environment. ### Does this PR introduce _any_ user-facing change? No ### How was this patch tested? By running the benchmark and checking its output. Closes #28440 from MaxGekk/SPARK-31527-DateTimeBenchmark-followup. Authored-by: Max Gekk <max.gekk@gmail.com> Signed-off-by: Takeshi Yamamuro <yamamuro@apache.org>
2020-05-03 20:39:50 -04:00
OpenJDK 64-Bit Server VM 11.0.7+10-post-Ubuntu-2ubuntu218.04 on Linux 4.15.0-1063-aws
Intel(R) Xeon(R) CPU E5-2670 v2 @ 2.50GHz
[SPARK-29320][TESTS] Compare `sql/core` module in JDK8/11 (Part 1) ### What changes were proposed in this pull request? This PR regenerates the `sql/core` benchmarks in JDK8/11 to compare the result. In general, we compare the ratio instead of the time. However, in this PR, the average time is compared. This PR should be considered as a rough comparison. **A. EXPECTED CASES(JDK11 is faster in general)** - [x] BloomFilterBenchmark (JDK11 is faster except one case) - [x] BuiltInDataSourceWriteBenchmark (JDK11 is faster at CSV/ORC) - [x] CSVBenchmark (JDK11 is faster except five cases) - [x] ColumnarBatchBenchmark (JDK11 is faster at `boolean`/`string` and some cases in `int`/`array`) - [x] DatasetBenchmark (JDK11 is faster with `string`, but is slower for `long` type) - [x] ExternalAppendOnlyUnsafeRowArrayBenchmark (JDK11 is faster except two cases) - [x] ExtractBenchmark (JDK11 is faster except HOUR/MINUTE/SECOND/MILLISECONDS/MICROSECONDS) - [x] HashedRelationMetricsBenchmark (JDK11 is faster) - [x] JSONBenchmark (JDK11 is much faster except eight cases) - [x] JoinBenchmark (JDK11 is faster except five cases) - [x] OrcNestedSchemaPruningBenchmark (JDK11 is faster in nine cases) - [x] PrimitiveArrayBenchmark (JDK11 is faster) - [x] SortBenchmark (JDK11 is faster except `Arrays.sort` case) - [x] UDFBenchmark (N/A, values are too small) - [x] UnsafeArrayDataBenchmark (JDK11 is faster except one case) - [x] WideTableBenchmark (JDK11 is faster except two cases) **B. CASES WE NEED TO INVESTIGATE MORE LATER** - [x] AggregateBenchmark (JDK11 is slower in general) - [x] CompressionSchemeBenchmark (JDK11 is slower in general except `string`) - [x] DataSourceReadBenchmark (JDK11 is slower in general) - [x] DateTimeBenchmark (JDK11 is slightly slower in general except `parsing`) - [x] MakeDateTimeBenchmark (JDK11 is slower except two cases) - [x] MiscBenchmark (JDK11 is slower except ten cases) - [x] OrcV2NestedSchemaPruningBenchmark (JDK11 is slower) - [x] ParquetNestedSchemaPruningBenchmark (JDK11 is slower except six cases) - [x] RangeBenchmark (JDK11 is slower except one case) `FilterPushdownBenchmark/InExpressionBenchmark/WideSchemaBenchmark` will be compared later because it took long timer. ### Why are the changes needed? According to the result, there are some difference between JDK8/JDK11. This will be a baseline for the future improvement and comparison. Also, as a reproducible environment, the following environment is used. - Instance: `r3.xlarge` - OS: `CentOS Linux release 7.5.1804 (Core)` - JDK: - `OpenJDK Runtime Environment (build 1.8.0_222-b10)` - `OpenJDK Runtime Environment 18.9 (build 11.0.4+11-LTS)` ### Does this PR introduce any user-facing change? No. ### How was this patch tested? This is a test-only PR. We need to run benchmark. Closes #26003 from dongjoon-hyun/SPARK-29320. Authored-by: Dongjoon Hyun <dhyun@apple.com> Signed-off-by: Dongjoon Hyun <dhyun@apple.com>
2019-10-03 11:58:25 -04:00
window: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative
------------------------------------------------------------------------------------------------------------------------
[SPARK-31630][SQL] Fix perf regression by skipping timestamps rebasing after some threshold ### What changes were proposed in this pull request? Skip timestamps rebasing after a global threshold when there is no difference between Julian and Gregorian calendars. This allows to avoid checking hash maps of switch points, and fixes perf regressions in `toJavaTimestamp()` and `fromJavaTimestamp()`. ### Why are the changes needed? The changes fix perf regressions of conversions to/from external type `java.sql.Timestamp`. Before (see the PR's results https://github.com/apache/spark/pull/28440): ``` ================================================================================================ Conversion from/to external types ================================================================================================ OpenJDK 64-Bit Server VM 1.8.0_252-8u252-b09-1~18.04-b09 on Linux 4.15.0-1063-aws Intel(R) Xeon(R) CPU E5-2670 v2 2.50GHz To/from Java's date-time: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative ------------------------------------------------------------------------------------------------------------------------ From java.sql.Timestamp 376 388 10 13.3 75.2 1.1X Collect java.sql.Timestamp 1878 1937 64 2.7 375.6 0.2X ``` After: ``` ================================================================================================ Conversion from/to external types ================================================================================================ OpenJDK 64-Bit Server VM 1.8.0_252-8u252-b09-1~18.04-b09 on Linux 4.15.0-1063-aws Intel(R) Xeon(R) CPU E5-2670 v2 2.50GHz To/from Java's date-time: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative ------------------------------------------------------------------------------------------------------------------------ From java.sql.Timestamp 249 264 24 20.1 49.8 1.7X Collect java.sql.Timestamp 1503 1523 24 3.3 300.5 0.3X ``` Perf improvements in average of: 1. From java.sql.Timestamp is ~ 34% 2. To java.sql.Timestamps is ~16% ### Does this PR introduce _any_ user-facing change? No ### How was this patch tested? By existing test suites `DateTimeUtilsSuite` and `RebaseDateTimeSuite`. Closes #28441 from MaxGekk/opt-rebase-common-threshold. Authored-by: Max Gekk <max.gekk@gmail.com> Signed-off-by: Wenchen Fan <wenchen@databricks.com>
2020-05-05 10:11:53 -04:00
window wholestage off 2044 2127 117 0.5 2044.3 1.0X
window wholestage on 48057 48109 54 0.0 48056.9 0.0X
[SPARK-29320][TESTS] Compare `sql/core` module in JDK8/11 (Part 1) ### What changes were proposed in this pull request? This PR regenerates the `sql/core` benchmarks in JDK8/11 to compare the result. In general, we compare the ratio instead of the time. However, in this PR, the average time is compared. This PR should be considered as a rough comparison. **A. EXPECTED CASES(JDK11 is faster in general)** - [x] BloomFilterBenchmark (JDK11 is faster except one case) - [x] BuiltInDataSourceWriteBenchmark (JDK11 is faster at CSV/ORC) - [x] CSVBenchmark (JDK11 is faster except five cases) - [x] ColumnarBatchBenchmark (JDK11 is faster at `boolean`/`string` and some cases in `int`/`array`) - [x] DatasetBenchmark (JDK11 is faster with `string`, but is slower for `long` type) - [x] ExternalAppendOnlyUnsafeRowArrayBenchmark (JDK11 is faster except two cases) - [x] ExtractBenchmark (JDK11 is faster except HOUR/MINUTE/SECOND/MILLISECONDS/MICROSECONDS) - [x] HashedRelationMetricsBenchmark (JDK11 is faster) - [x] JSONBenchmark (JDK11 is much faster except eight cases) - [x] JoinBenchmark (JDK11 is faster except five cases) - [x] OrcNestedSchemaPruningBenchmark (JDK11 is faster in nine cases) - [x] PrimitiveArrayBenchmark (JDK11 is faster) - [x] SortBenchmark (JDK11 is faster except `Arrays.sort` case) - [x] UDFBenchmark (N/A, values are too small) - [x] UnsafeArrayDataBenchmark (JDK11 is faster except one case) - [x] WideTableBenchmark (JDK11 is faster except two cases) **B. CASES WE NEED TO INVESTIGATE MORE LATER** - [x] AggregateBenchmark (JDK11 is slower in general) - [x] CompressionSchemeBenchmark (JDK11 is slower in general except `string`) - [x] DataSourceReadBenchmark (JDK11 is slower in general) - [x] DateTimeBenchmark (JDK11 is slightly slower in general except `parsing`) - [x] MakeDateTimeBenchmark (JDK11 is slower except two cases) - [x] MiscBenchmark (JDK11 is slower except ten cases) - [x] OrcV2NestedSchemaPruningBenchmark (JDK11 is slower) - [x] ParquetNestedSchemaPruningBenchmark (JDK11 is slower except six cases) - [x] RangeBenchmark (JDK11 is slower except one case) `FilterPushdownBenchmark/InExpressionBenchmark/WideSchemaBenchmark` will be compared later because it took long timer. ### Why are the changes needed? According to the result, there are some difference between JDK8/JDK11. This will be a baseline for the future improvement and comparison. Also, as a reproducible environment, the following environment is used. - Instance: `r3.xlarge` - OS: `CentOS Linux release 7.5.1804 (Core)` - JDK: - `OpenJDK Runtime Environment (build 1.8.0_222-b10)` - `OpenJDK Runtime Environment 18.9 (build 11.0.4+11-LTS)` ### Does this PR introduce any user-facing change? No. ### How was this patch tested? This is a test-only PR. We need to run benchmark. Closes #26003 from dongjoon-hyun/SPARK-29320. Authored-by: Dongjoon Hyun <dhyun@apple.com> Signed-off-by: Dongjoon Hyun <dhyun@apple.com>
2019-10-03 11:58:25 -04:00
================================================================================================
Truncation
================================================================================================
[SPARK-31527][SQL][TESTS][FOLLOWUP] Fix the number of rows in `DateTimeBenchmark` ### What changes were proposed in this pull request? - Changed to the number of rows in benchmark cases from 3 to the actual number `N`. - Regenerated benchmark results in the environment: | Item | Description | | ---- | ----| | Region | us-west-2 (Oregon) | | Instance | r3.xlarge | | AMI | ubuntu/images/hvm-ssd/ubuntu-bionic-18.04-amd64-server-20190722.1 (ami-06f2f779464715dc5) | | Java | OpenJDK 64-Bit Server VM 1.8.0_242 and OpenJDK 64-Bit Server VM 11.0.6+10 | ### Why are the changes needed? The changes are needed to have: - Correct benchmark results - Base line for other perf improvements that can be checked in the same environment. ### Does this PR introduce _any_ user-facing change? No ### How was this patch tested? By running the benchmark and checking its output. Closes #28440 from MaxGekk/SPARK-31527-DateTimeBenchmark-followup. Authored-by: Max Gekk <max.gekk@gmail.com> Signed-off-by: Takeshi Yamamuro <yamamuro@apache.org>
2020-05-03 20:39:50 -04:00
OpenJDK 64-Bit Server VM 11.0.7+10-post-Ubuntu-2ubuntu218.04 on Linux 4.15.0-1063-aws
Intel(R) Xeon(R) CPU E5-2670 v2 @ 2.50GHz
[SPARK-29320][TESTS] Compare `sql/core` module in JDK8/11 (Part 1) ### What changes were proposed in this pull request? This PR regenerates the `sql/core` benchmarks in JDK8/11 to compare the result. In general, we compare the ratio instead of the time. However, in this PR, the average time is compared. This PR should be considered as a rough comparison. **A. EXPECTED CASES(JDK11 is faster in general)** - [x] BloomFilterBenchmark (JDK11 is faster except one case) - [x] BuiltInDataSourceWriteBenchmark (JDK11 is faster at CSV/ORC) - [x] CSVBenchmark (JDK11 is faster except five cases) - [x] ColumnarBatchBenchmark (JDK11 is faster at `boolean`/`string` and some cases in `int`/`array`) - [x] DatasetBenchmark (JDK11 is faster with `string`, but is slower for `long` type) - [x] ExternalAppendOnlyUnsafeRowArrayBenchmark (JDK11 is faster except two cases) - [x] ExtractBenchmark (JDK11 is faster except HOUR/MINUTE/SECOND/MILLISECONDS/MICROSECONDS) - [x] HashedRelationMetricsBenchmark (JDK11 is faster) - [x] JSONBenchmark (JDK11 is much faster except eight cases) - [x] JoinBenchmark (JDK11 is faster except five cases) - [x] OrcNestedSchemaPruningBenchmark (JDK11 is faster in nine cases) - [x] PrimitiveArrayBenchmark (JDK11 is faster) - [x] SortBenchmark (JDK11 is faster except `Arrays.sort` case) - [x] UDFBenchmark (N/A, values are too small) - [x] UnsafeArrayDataBenchmark (JDK11 is faster except one case) - [x] WideTableBenchmark (JDK11 is faster except two cases) **B. CASES WE NEED TO INVESTIGATE MORE LATER** - [x] AggregateBenchmark (JDK11 is slower in general) - [x] CompressionSchemeBenchmark (JDK11 is slower in general except `string`) - [x] DataSourceReadBenchmark (JDK11 is slower in general) - [x] DateTimeBenchmark (JDK11 is slightly slower in general except `parsing`) - [x] MakeDateTimeBenchmark (JDK11 is slower except two cases) - [x] MiscBenchmark (JDK11 is slower except ten cases) - [x] OrcV2NestedSchemaPruningBenchmark (JDK11 is slower) - [x] ParquetNestedSchemaPruningBenchmark (JDK11 is slower except six cases) - [x] RangeBenchmark (JDK11 is slower except one case) `FilterPushdownBenchmark/InExpressionBenchmark/WideSchemaBenchmark` will be compared later because it took long timer. ### Why are the changes needed? According to the result, there are some difference between JDK8/JDK11. This will be a baseline for the future improvement and comparison. Also, as a reproducible environment, the following environment is used. - Instance: `r3.xlarge` - OS: `CentOS Linux release 7.5.1804 (Core)` - JDK: - `OpenJDK Runtime Environment (build 1.8.0_222-b10)` - `OpenJDK Runtime Environment 18.9 (build 11.0.4+11-LTS)` ### Does this PR introduce any user-facing change? No. ### How was this patch tested? This is a test-only PR. We need to run benchmark. Closes #26003 from dongjoon-hyun/SPARK-29320. Authored-by: Dongjoon Hyun <dhyun@apple.com> Signed-off-by: Dongjoon Hyun <dhyun@apple.com>
2019-10-03 11:58:25 -04:00
date_trunc YEAR: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative
------------------------------------------------------------------------------------------------------------------------
[SPARK-31630][SQL] Fix perf regression by skipping timestamps rebasing after some threshold ### What changes were proposed in this pull request? Skip timestamps rebasing after a global threshold when there is no difference between Julian and Gregorian calendars. This allows to avoid checking hash maps of switch points, and fixes perf regressions in `toJavaTimestamp()` and `fromJavaTimestamp()`. ### Why are the changes needed? The changes fix perf regressions of conversions to/from external type `java.sql.Timestamp`. Before (see the PR's results https://github.com/apache/spark/pull/28440): ``` ================================================================================================ Conversion from/to external types ================================================================================================ OpenJDK 64-Bit Server VM 1.8.0_252-8u252-b09-1~18.04-b09 on Linux 4.15.0-1063-aws Intel(R) Xeon(R) CPU E5-2670 v2 2.50GHz To/from Java's date-time: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative ------------------------------------------------------------------------------------------------------------------------ From java.sql.Timestamp 376 388 10 13.3 75.2 1.1X Collect java.sql.Timestamp 1878 1937 64 2.7 375.6 0.2X ``` After: ``` ================================================================================================ Conversion from/to external types ================================================================================================ OpenJDK 64-Bit Server VM 1.8.0_252-8u252-b09-1~18.04-b09 on Linux 4.15.0-1063-aws Intel(R) Xeon(R) CPU E5-2670 v2 2.50GHz To/from Java's date-time: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative ------------------------------------------------------------------------------------------------------------------------ From java.sql.Timestamp 249 264 24 20.1 49.8 1.7X Collect java.sql.Timestamp 1503 1523 24 3.3 300.5 0.3X ``` Perf improvements in average of: 1. From java.sql.Timestamp is ~ 34% 2. To java.sql.Timestamps is ~16% ### Does this PR introduce _any_ user-facing change? No ### How was this patch tested? By existing test suites `DateTimeUtilsSuite` and `RebaseDateTimeSuite`. Closes #28441 from MaxGekk/opt-rebase-common-threshold. Authored-by: Max Gekk <max.gekk@gmail.com> Signed-off-by: Wenchen Fan <wenchen@databricks.com>
2020-05-05 10:11:53 -04:00
date_trunc YEAR wholestage off 2540 2542 3 3.9 254.0 1.0X
date_trunc YEAR wholestage on 2486 2507 29 4.0 248.6 1.0X
[SPARK-29320][TESTS] Compare `sql/core` module in JDK8/11 (Part 1) ### What changes were proposed in this pull request? This PR regenerates the `sql/core` benchmarks in JDK8/11 to compare the result. In general, we compare the ratio instead of the time. However, in this PR, the average time is compared. This PR should be considered as a rough comparison. **A. EXPECTED CASES(JDK11 is faster in general)** - [x] BloomFilterBenchmark (JDK11 is faster except one case) - [x] BuiltInDataSourceWriteBenchmark (JDK11 is faster at CSV/ORC) - [x] CSVBenchmark (JDK11 is faster except five cases) - [x] ColumnarBatchBenchmark (JDK11 is faster at `boolean`/`string` and some cases in `int`/`array`) - [x] DatasetBenchmark (JDK11 is faster with `string`, but is slower for `long` type) - [x] ExternalAppendOnlyUnsafeRowArrayBenchmark (JDK11 is faster except two cases) - [x] ExtractBenchmark (JDK11 is faster except HOUR/MINUTE/SECOND/MILLISECONDS/MICROSECONDS) - [x] HashedRelationMetricsBenchmark (JDK11 is faster) - [x] JSONBenchmark (JDK11 is much faster except eight cases) - [x] JoinBenchmark (JDK11 is faster except five cases) - [x] OrcNestedSchemaPruningBenchmark (JDK11 is faster in nine cases) - [x] PrimitiveArrayBenchmark (JDK11 is faster) - [x] SortBenchmark (JDK11 is faster except `Arrays.sort` case) - [x] UDFBenchmark (N/A, values are too small) - [x] UnsafeArrayDataBenchmark (JDK11 is faster except one case) - [x] WideTableBenchmark (JDK11 is faster except two cases) **B. CASES WE NEED TO INVESTIGATE MORE LATER** - [x] AggregateBenchmark (JDK11 is slower in general) - [x] CompressionSchemeBenchmark (JDK11 is slower in general except `string`) - [x] DataSourceReadBenchmark (JDK11 is slower in general) - [x] DateTimeBenchmark (JDK11 is slightly slower in general except `parsing`) - [x] MakeDateTimeBenchmark (JDK11 is slower except two cases) - [x] MiscBenchmark (JDK11 is slower except ten cases) - [x] OrcV2NestedSchemaPruningBenchmark (JDK11 is slower) - [x] ParquetNestedSchemaPruningBenchmark (JDK11 is slower except six cases) - [x] RangeBenchmark (JDK11 is slower except one case) `FilterPushdownBenchmark/InExpressionBenchmark/WideSchemaBenchmark` will be compared later because it took long timer. ### Why are the changes needed? According to the result, there are some difference between JDK8/JDK11. This will be a baseline for the future improvement and comparison. Also, as a reproducible environment, the following environment is used. - Instance: `r3.xlarge` - OS: `CentOS Linux release 7.5.1804 (Core)` - JDK: - `OpenJDK Runtime Environment (build 1.8.0_222-b10)` - `OpenJDK Runtime Environment 18.9 (build 11.0.4+11-LTS)` ### Does this PR introduce any user-facing change? No. ### How was this patch tested? This is a test-only PR. We need to run benchmark. Closes #26003 from dongjoon-hyun/SPARK-29320. Authored-by: Dongjoon Hyun <dhyun@apple.com> Signed-off-by: Dongjoon Hyun <dhyun@apple.com>
2019-10-03 11:58:25 -04:00
[SPARK-31527][SQL][TESTS][FOLLOWUP] Fix the number of rows in `DateTimeBenchmark` ### What changes were proposed in this pull request? - Changed to the number of rows in benchmark cases from 3 to the actual number `N`. - Regenerated benchmark results in the environment: | Item | Description | | ---- | ----| | Region | us-west-2 (Oregon) | | Instance | r3.xlarge | | AMI | ubuntu/images/hvm-ssd/ubuntu-bionic-18.04-amd64-server-20190722.1 (ami-06f2f779464715dc5) | | Java | OpenJDK 64-Bit Server VM 1.8.0_242 and OpenJDK 64-Bit Server VM 11.0.6+10 | ### Why are the changes needed? The changes are needed to have: - Correct benchmark results - Base line for other perf improvements that can be checked in the same environment. ### Does this PR introduce _any_ user-facing change? No ### How was this patch tested? By running the benchmark and checking its output. Closes #28440 from MaxGekk/SPARK-31527-DateTimeBenchmark-followup. Authored-by: Max Gekk <max.gekk@gmail.com> Signed-off-by: Takeshi Yamamuro <yamamuro@apache.org>
2020-05-03 20:39:50 -04:00
OpenJDK 64-Bit Server VM 11.0.7+10-post-Ubuntu-2ubuntu218.04 on Linux 4.15.0-1063-aws
Intel(R) Xeon(R) CPU E5-2670 v2 @ 2.50GHz
[SPARK-29320][TESTS] Compare `sql/core` module in JDK8/11 (Part 1) ### What changes were proposed in this pull request? This PR regenerates the `sql/core` benchmarks in JDK8/11 to compare the result. In general, we compare the ratio instead of the time. However, in this PR, the average time is compared. This PR should be considered as a rough comparison. **A. EXPECTED CASES(JDK11 is faster in general)** - [x] BloomFilterBenchmark (JDK11 is faster except one case) - [x] BuiltInDataSourceWriteBenchmark (JDK11 is faster at CSV/ORC) - [x] CSVBenchmark (JDK11 is faster except five cases) - [x] ColumnarBatchBenchmark (JDK11 is faster at `boolean`/`string` and some cases in `int`/`array`) - [x] DatasetBenchmark (JDK11 is faster with `string`, but is slower for `long` type) - [x] ExternalAppendOnlyUnsafeRowArrayBenchmark (JDK11 is faster except two cases) - [x] ExtractBenchmark (JDK11 is faster except HOUR/MINUTE/SECOND/MILLISECONDS/MICROSECONDS) - [x] HashedRelationMetricsBenchmark (JDK11 is faster) - [x] JSONBenchmark (JDK11 is much faster except eight cases) - [x] JoinBenchmark (JDK11 is faster except five cases) - [x] OrcNestedSchemaPruningBenchmark (JDK11 is faster in nine cases) - [x] PrimitiveArrayBenchmark (JDK11 is faster) - [x] SortBenchmark (JDK11 is faster except `Arrays.sort` case) - [x] UDFBenchmark (N/A, values are too small) - [x] UnsafeArrayDataBenchmark (JDK11 is faster except one case) - [x] WideTableBenchmark (JDK11 is faster except two cases) **B. CASES WE NEED TO INVESTIGATE MORE LATER** - [x] AggregateBenchmark (JDK11 is slower in general) - [x] CompressionSchemeBenchmark (JDK11 is slower in general except `string`) - [x] DataSourceReadBenchmark (JDK11 is slower in general) - [x] DateTimeBenchmark (JDK11 is slightly slower in general except `parsing`) - [x] MakeDateTimeBenchmark (JDK11 is slower except two cases) - [x] MiscBenchmark (JDK11 is slower except ten cases) - [x] OrcV2NestedSchemaPruningBenchmark (JDK11 is slower) - [x] ParquetNestedSchemaPruningBenchmark (JDK11 is slower except six cases) - [x] RangeBenchmark (JDK11 is slower except one case) `FilterPushdownBenchmark/InExpressionBenchmark/WideSchemaBenchmark` will be compared later because it took long timer. ### Why are the changes needed? According to the result, there are some difference between JDK8/JDK11. This will be a baseline for the future improvement and comparison. Also, as a reproducible environment, the following environment is used. - Instance: `r3.xlarge` - OS: `CentOS Linux release 7.5.1804 (Core)` - JDK: - `OpenJDK Runtime Environment (build 1.8.0_222-b10)` - `OpenJDK Runtime Environment 18.9 (build 11.0.4+11-LTS)` ### Does this PR introduce any user-facing change? No. ### How was this patch tested? This is a test-only PR. We need to run benchmark. Closes #26003 from dongjoon-hyun/SPARK-29320. Authored-by: Dongjoon Hyun <dhyun@apple.com> Signed-off-by: Dongjoon Hyun <dhyun@apple.com>
2019-10-03 11:58:25 -04:00
date_trunc YYYY: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative
------------------------------------------------------------------------------------------------------------------------
[SPARK-31630][SQL] Fix perf regression by skipping timestamps rebasing after some threshold ### What changes were proposed in this pull request? Skip timestamps rebasing after a global threshold when there is no difference between Julian and Gregorian calendars. This allows to avoid checking hash maps of switch points, and fixes perf regressions in `toJavaTimestamp()` and `fromJavaTimestamp()`. ### Why are the changes needed? The changes fix perf regressions of conversions to/from external type `java.sql.Timestamp`. Before (see the PR's results https://github.com/apache/spark/pull/28440): ``` ================================================================================================ Conversion from/to external types ================================================================================================ OpenJDK 64-Bit Server VM 1.8.0_252-8u252-b09-1~18.04-b09 on Linux 4.15.0-1063-aws Intel(R) Xeon(R) CPU E5-2670 v2 2.50GHz To/from Java's date-time: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative ------------------------------------------------------------------------------------------------------------------------ From java.sql.Timestamp 376 388 10 13.3 75.2 1.1X Collect java.sql.Timestamp 1878 1937 64 2.7 375.6 0.2X ``` After: ``` ================================================================================================ Conversion from/to external types ================================================================================================ OpenJDK 64-Bit Server VM 1.8.0_252-8u252-b09-1~18.04-b09 on Linux 4.15.0-1063-aws Intel(R) Xeon(R) CPU E5-2670 v2 2.50GHz To/from Java's date-time: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative ------------------------------------------------------------------------------------------------------------------------ From java.sql.Timestamp 249 264 24 20.1 49.8 1.7X Collect java.sql.Timestamp 1503 1523 24 3.3 300.5 0.3X ``` Perf improvements in average of: 1. From java.sql.Timestamp is ~ 34% 2. To java.sql.Timestamps is ~16% ### Does this PR introduce _any_ user-facing change? No ### How was this patch tested? By existing test suites `DateTimeUtilsSuite` and `RebaseDateTimeSuite`. Closes #28441 from MaxGekk/opt-rebase-common-threshold. Authored-by: Max Gekk <max.gekk@gmail.com> Signed-off-by: Wenchen Fan <wenchen@databricks.com>
2020-05-05 10:11:53 -04:00
date_trunc YYYY wholestage off 2542 2543 3 3.9 254.2 1.0X
date_trunc YYYY wholestage on 2491 2498 9 4.0 249.1 1.0X
[SPARK-29320][TESTS] Compare `sql/core` module in JDK8/11 (Part 1) ### What changes were proposed in this pull request? This PR regenerates the `sql/core` benchmarks in JDK8/11 to compare the result. In general, we compare the ratio instead of the time. However, in this PR, the average time is compared. This PR should be considered as a rough comparison. **A. EXPECTED CASES(JDK11 is faster in general)** - [x] BloomFilterBenchmark (JDK11 is faster except one case) - [x] BuiltInDataSourceWriteBenchmark (JDK11 is faster at CSV/ORC) - [x] CSVBenchmark (JDK11 is faster except five cases) - [x] ColumnarBatchBenchmark (JDK11 is faster at `boolean`/`string` and some cases in `int`/`array`) - [x] DatasetBenchmark (JDK11 is faster with `string`, but is slower for `long` type) - [x] ExternalAppendOnlyUnsafeRowArrayBenchmark (JDK11 is faster except two cases) - [x] ExtractBenchmark (JDK11 is faster except HOUR/MINUTE/SECOND/MILLISECONDS/MICROSECONDS) - [x] HashedRelationMetricsBenchmark (JDK11 is faster) - [x] JSONBenchmark (JDK11 is much faster except eight cases) - [x] JoinBenchmark (JDK11 is faster except five cases) - [x] OrcNestedSchemaPruningBenchmark (JDK11 is faster in nine cases) - [x] PrimitiveArrayBenchmark (JDK11 is faster) - [x] SortBenchmark (JDK11 is faster except `Arrays.sort` case) - [x] UDFBenchmark (N/A, values are too small) - [x] UnsafeArrayDataBenchmark (JDK11 is faster except one case) - [x] WideTableBenchmark (JDK11 is faster except two cases) **B. CASES WE NEED TO INVESTIGATE MORE LATER** - [x] AggregateBenchmark (JDK11 is slower in general) - [x] CompressionSchemeBenchmark (JDK11 is slower in general except `string`) - [x] DataSourceReadBenchmark (JDK11 is slower in general) - [x] DateTimeBenchmark (JDK11 is slightly slower in general except `parsing`) - [x] MakeDateTimeBenchmark (JDK11 is slower except two cases) - [x] MiscBenchmark (JDK11 is slower except ten cases) - [x] OrcV2NestedSchemaPruningBenchmark (JDK11 is slower) - [x] ParquetNestedSchemaPruningBenchmark (JDK11 is slower except six cases) - [x] RangeBenchmark (JDK11 is slower except one case) `FilterPushdownBenchmark/InExpressionBenchmark/WideSchemaBenchmark` will be compared later because it took long timer. ### Why are the changes needed? According to the result, there are some difference between JDK8/JDK11. This will be a baseline for the future improvement and comparison. Also, as a reproducible environment, the following environment is used. - Instance: `r3.xlarge` - OS: `CentOS Linux release 7.5.1804 (Core)` - JDK: - `OpenJDK Runtime Environment (build 1.8.0_222-b10)` - `OpenJDK Runtime Environment 18.9 (build 11.0.4+11-LTS)` ### Does this PR introduce any user-facing change? No. ### How was this patch tested? This is a test-only PR. We need to run benchmark. Closes #26003 from dongjoon-hyun/SPARK-29320. Authored-by: Dongjoon Hyun <dhyun@apple.com> Signed-off-by: Dongjoon Hyun <dhyun@apple.com>
2019-10-03 11:58:25 -04:00
[SPARK-31527][SQL][TESTS][FOLLOWUP] Fix the number of rows in `DateTimeBenchmark` ### What changes were proposed in this pull request? - Changed to the number of rows in benchmark cases from 3 to the actual number `N`. - Regenerated benchmark results in the environment: | Item | Description | | ---- | ----| | Region | us-west-2 (Oregon) | | Instance | r3.xlarge | | AMI | ubuntu/images/hvm-ssd/ubuntu-bionic-18.04-amd64-server-20190722.1 (ami-06f2f779464715dc5) | | Java | OpenJDK 64-Bit Server VM 1.8.0_242 and OpenJDK 64-Bit Server VM 11.0.6+10 | ### Why are the changes needed? The changes are needed to have: - Correct benchmark results - Base line for other perf improvements that can be checked in the same environment. ### Does this PR introduce _any_ user-facing change? No ### How was this patch tested? By running the benchmark and checking its output. Closes #28440 from MaxGekk/SPARK-31527-DateTimeBenchmark-followup. Authored-by: Max Gekk <max.gekk@gmail.com> Signed-off-by: Takeshi Yamamuro <yamamuro@apache.org>
2020-05-03 20:39:50 -04:00
OpenJDK 64-Bit Server VM 11.0.7+10-post-Ubuntu-2ubuntu218.04 on Linux 4.15.0-1063-aws
Intel(R) Xeon(R) CPU E5-2670 v2 @ 2.50GHz
[SPARK-29320][TESTS] Compare `sql/core` module in JDK8/11 (Part 1) ### What changes were proposed in this pull request? This PR regenerates the `sql/core` benchmarks in JDK8/11 to compare the result. In general, we compare the ratio instead of the time. However, in this PR, the average time is compared. This PR should be considered as a rough comparison. **A. EXPECTED CASES(JDK11 is faster in general)** - [x] BloomFilterBenchmark (JDK11 is faster except one case) - [x] BuiltInDataSourceWriteBenchmark (JDK11 is faster at CSV/ORC) - [x] CSVBenchmark (JDK11 is faster except five cases) - [x] ColumnarBatchBenchmark (JDK11 is faster at `boolean`/`string` and some cases in `int`/`array`) - [x] DatasetBenchmark (JDK11 is faster with `string`, but is slower for `long` type) - [x] ExternalAppendOnlyUnsafeRowArrayBenchmark (JDK11 is faster except two cases) - [x] ExtractBenchmark (JDK11 is faster except HOUR/MINUTE/SECOND/MILLISECONDS/MICROSECONDS) - [x] HashedRelationMetricsBenchmark (JDK11 is faster) - [x] JSONBenchmark (JDK11 is much faster except eight cases) - [x] JoinBenchmark (JDK11 is faster except five cases) - [x] OrcNestedSchemaPruningBenchmark (JDK11 is faster in nine cases) - [x] PrimitiveArrayBenchmark (JDK11 is faster) - [x] SortBenchmark (JDK11 is faster except `Arrays.sort` case) - [x] UDFBenchmark (N/A, values are too small) - [x] UnsafeArrayDataBenchmark (JDK11 is faster except one case) - [x] WideTableBenchmark (JDK11 is faster except two cases) **B. CASES WE NEED TO INVESTIGATE MORE LATER** - [x] AggregateBenchmark (JDK11 is slower in general) - [x] CompressionSchemeBenchmark (JDK11 is slower in general except `string`) - [x] DataSourceReadBenchmark (JDK11 is slower in general) - [x] DateTimeBenchmark (JDK11 is slightly slower in general except `parsing`) - [x] MakeDateTimeBenchmark (JDK11 is slower except two cases) - [x] MiscBenchmark (JDK11 is slower except ten cases) - [x] OrcV2NestedSchemaPruningBenchmark (JDK11 is slower) - [x] ParquetNestedSchemaPruningBenchmark (JDK11 is slower except six cases) - [x] RangeBenchmark (JDK11 is slower except one case) `FilterPushdownBenchmark/InExpressionBenchmark/WideSchemaBenchmark` will be compared later because it took long timer. ### Why are the changes needed? According to the result, there are some difference between JDK8/JDK11. This will be a baseline for the future improvement and comparison. Also, as a reproducible environment, the following environment is used. - Instance: `r3.xlarge` - OS: `CentOS Linux release 7.5.1804 (Core)` - JDK: - `OpenJDK Runtime Environment (build 1.8.0_222-b10)` - `OpenJDK Runtime Environment 18.9 (build 11.0.4+11-LTS)` ### Does this PR introduce any user-facing change? No. ### How was this patch tested? This is a test-only PR. We need to run benchmark. Closes #26003 from dongjoon-hyun/SPARK-29320. Authored-by: Dongjoon Hyun <dhyun@apple.com> Signed-off-by: Dongjoon Hyun <dhyun@apple.com>
2019-10-03 11:58:25 -04:00
date_trunc YY: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative
------------------------------------------------------------------------------------------------------------------------
[SPARK-31630][SQL] Fix perf regression by skipping timestamps rebasing after some threshold ### What changes were proposed in this pull request? Skip timestamps rebasing after a global threshold when there is no difference between Julian and Gregorian calendars. This allows to avoid checking hash maps of switch points, and fixes perf regressions in `toJavaTimestamp()` and `fromJavaTimestamp()`. ### Why are the changes needed? The changes fix perf regressions of conversions to/from external type `java.sql.Timestamp`. Before (see the PR's results https://github.com/apache/spark/pull/28440): ``` ================================================================================================ Conversion from/to external types ================================================================================================ OpenJDK 64-Bit Server VM 1.8.0_252-8u252-b09-1~18.04-b09 on Linux 4.15.0-1063-aws Intel(R) Xeon(R) CPU E5-2670 v2 2.50GHz To/from Java's date-time: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative ------------------------------------------------------------------------------------------------------------------------ From java.sql.Timestamp 376 388 10 13.3 75.2 1.1X Collect java.sql.Timestamp 1878 1937 64 2.7 375.6 0.2X ``` After: ``` ================================================================================================ Conversion from/to external types ================================================================================================ OpenJDK 64-Bit Server VM 1.8.0_252-8u252-b09-1~18.04-b09 on Linux 4.15.0-1063-aws Intel(R) Xeon(R) CPU E5-2670 v2 2.50GHz To/from Java's date-time: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative ------------------------------------------------------------------------------------------------------------------------ From java.sql.Timestamp 249 264 24 20.1 49.8 1.7X Collect java.sql.Timestamp 1503 1523 24 3.3 300.5 0.3X ``` Perf improvements in average of: 1. From java.sql.Timestamp is ~ 34% 2. To java.sql.Timestamps is ~16% ### Does this PR introduce _any_ user-facing change? No ### How was this patch tested? By existing test suites `DateTimeUtilsSuite` and `RebaseDateTimeSuite`. Closes #28441 from MaxGekk/opt-rebase-common-threshold. Authored-by: Max Gekk <max.gekk@gmail.com> Signed-off-by: Wenchen Fan <wenchen@databricks.com>
2020-05-05 10:11:53 -04:00
date_trunc YY wholestage off 2545 2569 35 3.9 254.5 1.0X
date_trunc YY wholestage on 2487 2493 4 4.0 248.7 1.0X
[SPARK-29320][TESTS] Compare `sql/core` module in JDK8/11 (Part 1) ### What changes were proposed in this pull request? This PR regenerates the `sql/core` benchmarks in JDK8/11 to compare the result. In general, we compare the ratio instead of the time. However, in this PR, the average time is compared. This PR should be considered as a rough comparison. **A. EXPECTED CASES(JDK11 is faster in general)** - [x] BloomFilterBenchmark (JDK11 is faster except one case) - [x] BuiltInDataSourceWriteBenchmark (JDK11 is faster at CSV/ORC) - [x] CSVBenchmark (JDK11 is faster except five cases) - [x] ColumnarBatchBenchmark (JDK11 is faster at `boolean`/`string` and some cases in `int`/`array`) - [x] DatasetBenchmark (JDK11 is faster with `string`, but is slower for `long` type) - [x] ExternalAppendOnlyUnsafeRowArrayBenchmark (JDK11 is faster except two cases) - [x] ExtractBenchmark (JDK11 is faster except HOUR/MINUTE/SECOND/MILLISECONDS/MICROSECONDS) - [x] HashedRelationMetricsBenchmark (JDK11 is faster) - [x] JSONBenchmark (JDK11 is much faster except eight cases) - [x] JoinBenchmark (JDK11 is faster except five cases) - [x] OrcNestedSchemaPruningBenchmark (JDK11 is faster in nine cases) - [x] PrimitiveArrayBenchmark (JDK11 is faster) - [x] SortBenchmark (JDK11 is faster except `Arrays.sort` case) - [x] UDFBenchmark (N/A, values are too small) - [x] UnsafeArrayDataBenchmark (JDK11 is faster except one case) - [x] WideTableBenchmark (JDK11 is faster except two cases) **B. CASES WE NEED TO INVESTIGATE MORE LATER** - [x] AggregateBenchmark (JDK11 is slower in general) - [x] CompressionSchemeBenchmark (JDK11 is slower in general except `string`) - [x] DataSourceReadBenchmark (JDK11 is slower in general) - [x] DateTimeBenchmark (JDK11 is slightly slower in general except `parsing`) - [x] MakeDateTimeBenchmark (JDK11 is slower except two cases) - [x] MiscBenchmark (JDK11 is slower except ten cases) - [x] OrcV2NestedSchemaPruningBenchmark (JDK11 is slower) - [x] ParquetNestedSchemaPruningBenchmark (JDK11 is slower except six cases) - [x] RangeBenchmark (JDK11 is slower except one case) `FilterPushdownBenchmark/InExpressionBenchmark/WideSchemaBenchmark` will be compared later because it took long timer. ### Why are the changes needed? According to the result, there are some difference between JDK8/JDK11. This will be a baseline for the future improvement and comparison. Also, as a reproducible environment, the following environment is used. - Instance: `r3.xlarge` - OS: `CentOS Linux release 7.5.1804 (Core)` - JDK: - `OpenJDK Runtime Environment (build 1.8.0_222-b10)` - `OpenJDK Runtime Environment 18.9 (build 11.0.4+11-LTS)` ### Does this PR introduce any user-facing change? No. ### How was this patch tested? This is a test-only PR. We need to run benchmark. Closes #26003 from dongjoon-hyun/SPARK-29320. Authored-by: Dongjoon Hyun <dhyun@apple.com> Signed-off-by: Dongjoon Hyun <dhyun@apple.com>
2019-10-03 11:58:25 -04:00
[SPARK-31527][SQL][TESTS][FOLLOWUP] Fix the number of rows in `DateTimeBenchmark` ### What changes were proposed in this pull request? - Changed to the number of rows in benchmark cases from 3 to the actual number `N`. - Regenerated benchmark results in the environment: | Item | Description | | ---- | ----| | Region | us-west-2 (Oregon) | | Instance | r3.xlarge | | AMI | ubuntu/images/hvm-ssd/ubuntu-bionic-18.04-amd64-server-20190722.1 (ami-06f2f779464715dc5) | | Java | OpenJDK 64-Bit Server VM 1.8.0_242 and OpenJDK 64-Bit Server VM 11.0.6+10 | ### Why are the changes needed? The changes are needed to have: - Correct benchmark results - Base line for other perf improvements that can be checked in the same environment. ### Does this PR introduce _any_ user-facing change? No ### How was this patch tested? By running the benchmark and checking its output. Closes #28440 from MaxGekk/SPARK-31527-DateTimeBenchmark-followup. Authored-by: Max Gekk <max.gekk@gmail.com> Signed-off-by: Takeshi Yamamuro <yamamuro@apache.org>
2020-05-03 20:39:50 -04:00
OpenJDK 64-Bit Server VM 11.0.7+10-post-Ubuntu-2ubuntu218.04 on Linux 4.15.0-1063-aws
Intel(R) Xeon(R) CPU E5-2670 v2 @ 2.50GHz
[SPARK-29320][TESTS] Compare `sql/core` module in JDK8/11 (Part 1) ### What changes were proposed in this pull request? This PR regenerates the `sql/core` benchmarks in JDK8/11 to compare the result. In general, we compare the ratio instead of the time. However, in this PR, the average time is compared. This PR should be considered as a rough comparison. **A. EXPECTED CASES(JDK11 is faster in general)** - [x] BloomFilterBenchmark (JDK11 is faster except one case) - [x] BuiltInDataSourceWriteBenchmark (JDK11 is faster at CSV/ORC) - [x] CSVBenchmark (JDK11 is faster except five cases) - [x] ColumnarBatchBenchmark (JDK11 is faster at `boolean`/`string` and some cases in `int`/`array`) - [x] DatasetBenchmark (JDK11 is faster with `string`, but is slower for `long` type) - [x] ExternalAppendOnlyUnsafeRowArrayBenchmark (JDK11 is faster except two cases) - [x] ExtractBenchmark (JDK11 is faster except HOUR/MINUTE/SECOND/MILLISECONDS/MICROSECONDS) - [x] HashedRelationMetricsBenchmark (JDK11 is faster) - [x] JSONBenchmark (JDK11 is much faster except eight cases) - [x] JoinBenchmark (JDK11 is faster except five cases) - [x] OrcNestedSchemaPruningBenchmark (JDK11 is faster in nine cases) - [x] PrimitiveArrayBenchmark (JDK11 is faster) - [x] SortBenchmark (JDK11 is faster except `Arrays.sort` case) - [x] UDFBenchmark (N/A, values are too small) - [x] UnsafeArrayDataBenchmark (JDK11 is faster except one case) - [x] WideTableBenchmark (JDK11 is faster except two cases) **B. CASES WE NEED TO INVESTIGATE MORE LATER** - [x] AggregateBenchmark (JDK11 is slower in general) - [x] CompressionSchemeBenchmark (JDK11 is slower in general except `string`) - [x] DataSourceReadBenchmark (JDK11 is slower in general) - [x] DateTimeBenchmark (JDK11 is slightly slower in general except `parsing`) - [x] MakeDateTimeBenchmark (JDK11 is slower except two cases) - [x] MiscBenchmark (JDK11 is slower except ten cases) - [x] OrcV2NestedSchemaPruningBenchmark (JDK11 is slower) - [x] ParquetNestedSchemaPruningBenchmark (JDK11 is slower except six cases) - [x] RangeBenchmark (JDK11 is slower except one case) `FilterPushdownBenchmark/InExpressionBenchmark/WideSchemaBenchmark` will be compared later because it took long timer. ### Why are the changes needed? According to the result, there are some difference between JDK8/JDK11. This will be a baseline for the future improvement and comparison. Also, as a reproducible environment, the following environment is used. - Instance: `r3.xlarge` - OS: `CentOS Linux release 7.5.1804 (Core)` - JDK: - `OpenJDK Runtime Environment (build 1.8.0_222-b10)` - `OpenJDK Runtime Environment 18.9 (build 11.0.4+11-LTS)` ### Does this PR introduce any user-facing change? No. ### How was this patch tested? This is a test-only PR. We need to run benchmark. Closes #26003 from dongjoon-hyun/SPARK-29320. Authored-by: Dongjoon Hyun <dhyun@apple.com> Signed-off-by: Dongjoon Hyun <dhyun@apple.com>
2019-10-03 11:58:25 -04:00
date_trunc MON: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative
------------------------------------------------------------------------------------------------------------------------
[SPARK-31630][SQL] Fix perf regression by skipping timestamps rebasing after some threshold ### What changes were proposed in this pull request? Skip timestamps rebasing after a global threshold when there is no difference between Julian and Gregorian calendars. This allows to avoid checking hash maps of switch points, and fixes perf regressions in `toJavaTimestamp()` and `fromJavaTimestamp()`. ### Why are the changes needed? The changes fix perf regressions of conversions to/from external type `java.sql.Timestamp`. Before (see the PR's results https://github.com/apache/spark/pull/28440): ``` ================================================================================================ Conversion from/to external types ================================================================================================ OpenJDK 64-Bit Server VM 1.8.0_252-8u252-b09-1~18.04-b09 on Linux 4.15.0-1063-aws Intel(R) Xeon(R) CPU E5-2670 v2 2.50GHz To/from Java's date-time: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative ------------------------------------------------------------------------------------------------------------------------ From java.sql.Timestamp 376 388 10 13.3 75.2 1.1X Collect java.sql.Timestamp 1878 1937 64 2.7 375.6 0.2X ``` After: ``` ================================================================================================ Conversion from/to external types ================================================================================================ OpenJDK 64-Bit Server VM 1.8.0_252-8u252-b09-1~18.04-b09 on Linux 4.15.0-1063-aws Intel(R) Xeon(R) CPU E5-2670 v2 2.50GHz To/from Java's date-time: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative ------------------------------------------------------------------------------------------------------------------------ From java.sql.Timestamp 249 264 24 20.1 49.8 1.7X Collect java.sql.Timestamp 1503 1523 24 3.3 300.5 0.3X ``` Perf improvements in average of: 1. From java.sql.Timestamp is ~ 34% 2. To java.sql.Timestamps is ~16% ### Does this PR introduce _any_ user-facing change? No ### How was this patch tested? By existing test suites `DateTimeUtilsSuite` and `RebaseDateTimeSuite`. Closes #28441 from MaxGekk/opt-rebase-common-threshold. Authored-by: Max Gekk <max.gekk@gmail.com> Signed-off-by: Wenchen Fan <wenchen@databricks.com>
2020-05-05 10:11:53 -04:00
date_trunc MON wholestage off 2590 2590 1 3.9 259.0 1.0X
date_trunc MON wholestage on 2506 2520 12 4.0 250.6 1.0X
[SPARK-29320][TESTS] Compare `sql/core` module in JDK8/11 (Part 1) ### What changes were proposed in this pull request? This PR regenerates the `sql/core` benchmarks in JDK8/11 to compare the result. In general, we compare the ratio instead of the time. However, in this PR, the average time is compared. This PR should be considered as a rough comparison. **A. EXPECTED CASES(JDK11 is faster in general)** - [x] BloomFilterBenchmark (JDK11 is faster except one case) - [x] BuiltInDataSourceWriteBenchmark (JDK11 is faster at CSV/ORC) - [x] CSVBenchmark (JDK11 is faster except five cases) - [x] ColumnarBatchBenchmark (JDK11 is faster at `boolean`/`string` and some cases in `int`/`array`) - [x] DatasetBenchmark (JDK11 is faster with `string`, but is slower for `long` type) - [x] ExternalAppendOnlyUnsafeRowArrayBenchmark (JDK11 is faster except two cases) - [x] ExtractBenchmark (JDK11 is faster except HOUR/MINUTE/SECOND/MILLISECONDS/MICROSECONDS) - [x] HashedRelationMetricsBenchmark (JDK11 is faster) - [x] JSONBenchmark (JDK11 is much faster except eight cases) - [x] JoinBenchmark (JDK11 is faster except five cases) - [x] OrcNestedSchemaPruningBenchmark (JDK11 is faster in nine cases) - [x] PrimitiveArrayBenchmark (JDK11 is faster) - [x] SortBenchmark (JDK11 is faster except `Arrays.sort` case) - [x] UDFBenchmark (N/A, values are too small) - [x] UnsafeArrayDataBenchmark (JDK11 is faster except one case) - [x] WideTableBenchmark (JDK11 is faster except two cases) **B. CASES WE NEED TO INVESTIGATE MORE LATER** - [x] AggregateBenchmark (JDK11 is slower in general) - [x] CompressionSchemeBenchmark (JDK11 is slower in general except `string`) - [x] DataSourceReadBenchmark (JDK11 is slower in general) - [x] DateTimeBenchmark (JDK11 is slightly slower in general except `parsing`) - [x] MakeDateTimeBenchmark (JDK11 is slower except two cases) - [x] MiscBenchmark (JDK11 is slower except ten cases) - [x] OrcV2NestedSchemaPruningBenchmark (JDK11 is slower) - [x] ParquetNestedSchemaPruningBenchmark (JDK11 is slower except six cases) - [x] RangeBenchmark (JDK11 is slower except one case) `FilterPushdownBenchmark/InExpressionBenchmark/WideSchemaBenchmark` will be compared later because it took long timer. ### Why are the changes needed? According to the result, there are some difference between JDK8/JDK11. This will be a baseline for the future improvement and comparison. Also, as a reproducible environment, the following environment is used. - Instance: `r3.xlarge` - OS: `CentOS Linux release 7.5.1804 (Core)` - JDK: - `OpenJDK Runtime Environment (build 1.8.0_222-b10)` - `OpenJDK Runtime Environment 18.9 (build 11.0.4+11-LTS)` ### Does this PR introduce any user-facing change? No. ### How was this patch tested? This is a test-only PR. We need to run benchmark. Closes #26003 from dongjoon-hyun/SPARK-29320. Authored-by: Dongjoon Hyun <dhyun@apple.com> Signed-off-by: Dongjoon Hyun <dhyun@apple.com>
2019-10-03 11:58:25 -04:00
[SPARK-31527][SQL][TESTS][FOLLOWUP] Fix the number of rows in `DateTimeBenchmark` ### What changes were proposed in this pull request? - Changed to the number of rows in benchmark cases from 3 to the actual number `N`. - Regenerated benchmark results in the environment: | Item | Description | | ---- | ----| | Region | us-west-2 (Oregon) | | Instance | r3.xlarge | | AMI | ubuntu/images/hvm-ssd/ubuntu-bionic-18.04-amd64-server-20190722.1 (ami-06f2f779464715dc5) | | Java | OpenJDK 64-Bit Server VM 1.8.0_242 and OpenJDK 64-Bit Server VM 11.0.6+10 | ### Why are the changes needed? The changes are needed to have: - Correct benchmark results - Base line for other perf improvements that can be checked in the same environment. ### Does this PR introduce _any_ user-facing change? No ### How was this patch tested? By running the benchmark and checking its output. Closes #28440 from MaxGekk/SPARK-31527-DateTimeBenchmark-followup. Authored-by: Max Gekk <max.gekk@gmail.com> Signed-off-by: Takeshi Yamamuro <yamamuro@apache.org>
2020-05-03 20:39:50 -04:00
OpenJDK 64-Bit Server VM 11.0.7+10-post-Ubuntu-2ubuntu218.04 on Linux 4.15.0-1063-aws
Intel(R) Xeon(R) CPU E5-2670 v2 @ 2.50GHz
[SPARK-29320][TESTS] Compare `sql/core` module in JDK8/11 (Part 1) ### What changes were proposed in this pull request? This PR regenerates the `sql/core` benchmarks in JDK8/11 to compare the result. In general, we compare the ratio instead of the time. However, in this PR, the average time is compared. This PR should be considered as a rough comparison. **A. EXPECTED CASES(JDK11 is faster in general)** - [x] BloomFilterBenchmark (JDK11 is faster except one case) - [x] BuiltInDataSourceWriteBenchmark (JDK11 is faster at CSV/ORC) - [x] CSVBenchmark (JDK11 is faster except five cases) - [x] ColumnarBatchBenchmark (JDK11 is faster at `boolean`/`string` and some cases in `int`/`array`) - [x] DatasetBenchmark (JDK11 is faster with `string`, but is slower for `long` type) - [x] ExternalAppendOnlyUnsafeRowArrayBenchmark (JDK11 is faster except two cases) - [x] ExtractBenchmark (JDK11 is faster except HOUR/MINUTE/SECOND/MILLISECONDS/MICROSECONDS) - [x] HashedRelationMetricsBenchmark (JDK11 is faster) - [x] JSONBenchmark (JDK11 is much faster except eight cases) - [x] JoinBenchmark (JDK11 is faster except five cases) - [x] OrcNestedSchemaPruningBenchmark (JDK11 is faster in nine cases) - [x] PrimitiveArrayBenchmark (JDK11 is faster) - [x] SortBenchmark (JDK11 is faster except `Arrays.sort` case) - [x] UDFBenchmark (N/A, values are too small) - [x] UnsafeArrayDataBenchmark (JDK11 is faster except one case) - [x] WideTableBenchmark (JDK11 is faster except two cases) **B. CASES WE NEED TO INVESTIGATE MORE LATER** - [x] AggregateBenchmark (JDK11 is slower in general) - [x] CompressionSchemeBenchmark (JDK11 is slower in general except `string`) - [x] DataSourceReadBenchmark (JDK11 is slower in general) - [x] DateTimeBenchmark (JDK11 is slightly slower in general except `parsing`) - [x] MakeDateTimeBenchmark (JDK11 is slower except two cases) - [x] MiscBenchmark (JDK11 is slower except ten cases) - [x] OrcV2NestedSchemaPruningBenchmark (JDK11 is slower) - [x] ParquetNestedSchemaPruningBenchmark (JDK11 is slower except six cases) - [x] RangeBenchmark (JDK11 is slower except one case) `FilterPushdownBenchmark/InExpressionBenchmark/WideSchemaBenchmark` will be compared later because it took long timer. ### Why are the changes needed? According to the result, there are some difference between JDK8/JDK11. This will be a baseline for the future improvement and comparison. Also, as a reproducible environment, the following environment is used. - Instance: `r3.xlarge` - OS: `CentOS Linux release 7.5.1804 (Core)` - JDK: - `OpenJDK Runtime Environment (build 1.8.0_222-b10)` - `OpenJDK Runtime Environment 18.9 (build 11.0.4+11-LTS)` ### Does this PR introduce any user-facing change? No. ### How was this patch tested? This is a test-only PR. We need to run benchmark. Closes #26003 from dongjoon-hyun/SPARK-29320. Authored-by: Dongjoon Hyun <dhyun@apple.com> Signed-off-by: Dongjoon Hyun <dhyun@apple.com>
2019-10-03 11:58:25 -04:00
date_trunc MONTH: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative
------------------------------------------------------------------------------------------------------------------------
[SPARK-31630][SQL] Fix perf regression by skipping timestamps rebasing after some threshold ### What changes were proposed in this pull request? Skip timestamps rebasing after a global threshold when there is no difference between Julian and Gregorian calendars. This allows to avoid checking hash maps of switch points, and fixes perf regressions in `toJavaTimestamp()` and `fromJavaTimestamp()`. ### Why are the changes needed? The changes fix perf regressions of conversions to/from external type `java.sql.Timestamp`. Before (see the PR's results https://github.com/apache/spark/pull/28440): ``` ================================================================================================ Conversion from/to external types ================================================================================================ OpenJDK 64-Bit Server VM 1.8.0_252-8u252-b09-1~18.04-b09 on Linux 4.15.0-1063-aws Intel(R) Xeon(R) CPU E5-2670 v2 2.50GHz To/from Java's date-time: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative ------------------------------------------------------------------------------------------------------------------------ From java.sql.Timestamp 376 388 10 13.3 75.2 1.1X Collect java.sql.Timestamp 1878 1937 64 2.7 375.6 0.2X ``` After: ``` ================================================================================================ Conversion from/to external types ================================================================================================ OpenJDK 64-Bit Server VM 1.8.0_252-8u252-b09-1~18.04-b09 on Linux 4.15.0-1063-aws Intel(R) Xeon(R) CPU E5-2670 v2 2.50GHz To/from Java's date-time: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative ------------------------------------------------------------------------------------------------------------------------ From java.sql.Timestamp 249 264 24 20.1 49.8 1.7X Collect java.sql.Timestamp 1503 1523 24 3.3 300.5 0.3X ``` Perf improvements in average of: 1. From java.sql.Timestamp is ~ 34% 2. To java.sql.Timestamps is ~16% ### Does this PR introduce _any_ user-facing change? No ### How was this patch tested? By existing test suites `DateTimeUtilsSuite` and `RebaseDateTimeSuite`. Closes #28441 from MaxGekk/opt-rebase-common-threshold. Authored-by: Max Gekk <max.gekk@gmail.com> Signed-off-by: Wenchen Fan <wenchen@databricks.com>
2020-05-05 10:11:53 -04:00
date_trunc MONTH wholestage off 2595 2603 11 3.9 259.5 1.0X
date_trunc MONTH wholestage on 2505 2516 12 4.0 250.5 1.0X
[SPARK-29320][TESTS] Compare `sql/core` module in JDK8/11 (Part 1) ### What changes were proposed in this pull request? This PR regenerates the `sql/core` benchmarks in JDK8/11 to compare the result. In general, we compare the ratio instead of the time. However, in this PR, the average time is compared. This PR should be considered as a rough comparison. **A. EXPECTED CASES(JDK11 is faster in general)** - [x] BloomFilterBenchmark (JDK11 is faster except one case) - [x] BuiltInDataSourceWriteBenchmark (JDK11 is faster at CSV/ORC) - [x] CSVBenchmark (JDK11 is faster except five cases) - [x] ColumnarBatchBenchmark (JDK11 is faster at `boolean`/`string` and some cases in `int`/`array`) - [x] DatasetBenchmark (JDK11 is faster with `string`, but is slower for `long` type) - [x] ExternalAppendOnlyUnsafeRowArrayBenchmark (JDK11 is faster except two cases) - [x] ExtractBenchmark (JDK11 is faster except HOUR/MINUTE/SECOND/MILLISECONDS/MICROSECONDS) - [x] HashedRelationMetricsBenchmark (JDK11 is faster) - [x] JSONBenchmark (JDK11 is much faster except eight cases) - [x] JoinBenchmark (JDK11 is faster except five cases) - [x] OrcNestedSchemaPruningBenchmark (JDK11 is faster in nine cases) - [x] PrimitiveArrayBenchmark (JDK11 is faster) - [x] SortBenchmark (JDK11 is faster except `Arrays.sort` case) - [x] UDFBenchmark (N/A, values are too small) - [x] UnsafeArrayDataBenchmark (JDK11 is faster except one case) - [x] WideTableBenchmark (JDK11 is faster except two cases) **B. CASES WE NEED TO INVESTIGATE MORE LATER** - [x] AggregateBenchmark (JDK11 is slower in general) - [x] CompressionSchemeBenchmark (JDK11 is slower in general except `string`) - [x] DataSourceReadBenchmark (JDK11 is slower in general) - [x] DateTimeBenchmark (JDK11 is slightly slower in general except `parsing`) - [x] MakeDateTimeBenchmark (JDK11 is slower except two cases) - [x] MiscBenchmark (JDK11 is slower except ten cases) - [x] OrcV2NestedSchemaPruningBenchmark (JDK11 is slower) - [x] ParquetNestedSchemaPruningBenchmark (JDK11 is slower except six cases) - [x] RangeBenchmark (JDK11 is slower except one case) `FilterPushdownBenchmark/InExpressionBenchmark/WideSchemaBenchmark` will be compared later because it took long timer. ### Why are the changes needed? According to the result, there are some difference between JDK8/JDK11. This will be a baseline for the future improvement and comparison. Also, as a reproducible environment, the following environment is used. - Instance: `r3.xlarge` - OS: `CentOS Linux release 7.5.1804 (Core)` - JDK: - `OpenJDK Runtime Environment (build 1.8.0_222-b10)` - `OpenJDK Runtime Environment 18.9 (build 11.0.4+11-LTS)` ### Does this PR introduce any user-facing change? No. ### How was this patch tested? This is a test-only PR. We need to run benchmark. Closes #26003 from dongjoon-hyun/SPARK-29320. Authored-by: Dongjoon Hyun <dhyun@apple.com> Signed-off-by: Dongjoon Hyun <dhyun@apple.com>
2019-10-03 11:58:25 -04:00
[SPARK-31527][SQL][TESTS][FOLLOWUP] Fix the number of rows in `DateTimeBenchmark` ### What changes were proposed in this pull request? - Changed to the number of rows in benchmark cases from 3 to the actual number `N`. - Regenerated benchmark results in the environment: | Item | Description | | ---- | ----| | Region | us-west-2 (Oregon) | | Instance | r3.xlarge | | AMI | ubuntu/images/hvm-ssd/ubuntu-bionic-18.04-amd64-server-20190722.1 (ami-06f2f779464715dc5) | | Java | OpenJDK 64-Bit Server VM 1.8.0_242 and OpenJDK 64-Bit Server VM 11.0.6+10 | ### Why are the changes needed? The changes are needed to have: - Correct benchmark results - Base line for other perf improvements that can be checked in the same environment. ### Does this PR introduce _any_ user-facing change? No ### How was this patch tested? By running the benchmark and checking its output. Closes #28440 from MaxGekk/SPARK-31527-DateTimeBenchmark-followup. Authored-by: Max Gekk <max.gekk@gmail.com> Signed-off-by: Takeshi Yamamuro <yamamuro@apache.org>
2020-05-03 20:39:50 -04:00
OpenJDK 64-Bit Server VM 11.0.7+10-post-Ubuntu-2ubuntu218.04 on Linux 4.15.0-1063-aws
Intel(R) Xeon(R) CPU E5-2670 v2 @ 2.50GHz
[SPARK-29320][TESTS] Compare `sql/core` module in JDK8/11 (Part 1) ### What changes were proposed in this pull request? This PR regenerates the `sql/core` benchmarks in JDK8/11 to compare the result. In general, we compare the ratio instead of the time. However, in this PR, the average time is compared. This PR should be considered as a rough comparison. **A. EXPECTED CASES(JDK11 is faster in general)** - [x] BloomFilterBenchmark (JDK11 is faster except one case) - [x] BuiltInDataSourceWriteBenchmark (JDK11 is faster at CSV/ORC) - [x] CSVBenchmark (JDK11 is faster except five cases) - [x] ColumnarBatchBenchmark (JDK11 is faster at `boolean`/`string` and some cases in `int`/`array`) - [x] DatasetBenchmark (JDK11 is faster with `string`, but is slower for `long` type) - [x] ExternalAppendOnlyUnsafeRowArrayBenchmark (JDK11 is faster except two cases) - [x] ExtractBenchmark (JDK11 is faster except HOUR/MINUTE/SECOND/MILLISECONDS/MICROSECONDS) - [x] HashedRelationMetricsBenchmark (JDK11 is faster) - [x] JSONBenchmark (JDK11 is much faster except eight cases) - [x] JoinBenchmark (JDK11 is faster except five cases) - [x] OrcNestedSchemaPruningBenchmark (JDK11 is faster in nine cases) - [x] PrimitiveArrayBenchmark (JDK11 is faster) - [x] SortBenchmark (JDK11 is faster except `Arrays.sort` case) - [x] UDFBenchmark (N/A, values are too small) - [x] UnsafeArrayDataBenchmark (JDK11 is faster except one case) - [x] WideTableBenchmark (JDK11 is faster except two cases) **B. CASES WE NEED TO INVESTIGATE MORE LATER** - [x] AggregateBenchmark (JDK11 is slower in general) - [x] CompressionSchemeBenchmark (JDK11 is slower in general except `string`) - [x] DataSourceReadBenchmark (JDK11 is slower in general) - [x] DateTimeBenchmark (JDK11 is slightly slower in general except `parsing`) - [x] MakeDateTimeBenchmark (JDK11 is slower except two cases) - [x] MiscBenchmark (JDK11 is slower except ten cases) - [x] OrcV2NestedSchemaPruningBenchmark (JDK11 is slower) - [x] ParquetNestedSchemaPruningBenchmark (JDK11 is slower except six cases) - [x] RangeBenchmark (JDK11 is slower except one case) `FilterPushdownBenchmark/InExpressionBenchmark/WideSchemaBenchmark` will be compared later because it took long timer. ### Why are the changes needed? According to the result, there are some difference between JDK8/JDK11. This will be a baseline for the future improvement and comparison. Also, as a reproducible environment, the following environment is used. - Instance: `r3.xlarge` - OS: `CentOS Linux release 7.5.1804 (Core)` - JDK: - `OpenJDK Runtime Environment (build 1.8.0_222-b10)` - `OpenJDK Runtime Environment 18.9 (build 11.0.4+11-LTS)` ### Does this PR introduce any user-facing change? No. ### How was this patch tested? This is a test-only PR. We need to run benchmark. Closes #26003 from dongjoon-hyun/SPARK-29320. Authored-by: Dongjoon Hyun <dhyun@apple.com> Signed-off-by: Dongjoon Hyun <dhyun@apple.com>
2019-10-03 11:58:25 -04:00
date_trunc MM: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative
------------------------------------------------------------------------------------------------------------------------
[SPARK-31630][SQL] Fix perf regression by skipping timestamps rebasing after some threshold ### What changes were proposed in this pull request? Skip timestamps rebasing after a global threshold when there is no difference between Julian and Gregorian calendars. This allows to avoid checking hash maps of switch points, and fixes perf regressions in `toJavaTimestamp()` and `fromJavaTimestamp()`. ### Why are the changes needed? The changes fix perf regressions of conversions to/from external type `java.sql.Timestamp`. Before (see the PR's results https://github.com/apache/spark/pull/28440): ``` ================================================================================================ Conversion from/to external types ================================================================================================ OpenJDK 64-Bit Server VM 1.8.0_252-8u252-b09-1~18.04-b09 on Linux 4.15.0-1063-aws Intel(R) Xeon(R) CPU E5-2670 v2 2.50GHz To/from Java's date-time: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative ------------------------------------------------------------------------------------------------------------------------ From java.sql.Timestamp 376 388 10 13.3 75.2 1.1X Collect java.sql.Timestamp 1878 1937 64 2.7 375.6 0.2X ``` After: ``` ================================================================================================ Conversion from/to external types ================================================================================================ OpenJDK 64-Bit Server VM 1.8.0_252-8u252-b09-1~18.04-b09 on Linux 4.15.0-1063-aws Intel(R) Xeon(R) CPU E5-2670 v2 2.50GHz To/from Java's date-time: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative ------------------------------------------------------------------------------------------------------------------------ From java.sql.Timestamp 249 264 24 20.1 49.8 1.7X Collect java.sql.Timestamp 1503 1523 24 3.3 300.5 0.3X ``` Perf improvements in average of: 1. From java.sql.Timestamp is ~ 34% 2. To java.sql.Timestamps is ~16% ### Does this PR introduce _any_ user-facing change? No ### How was this patch tested? By existing test suites `DateTimeUtilsSuite` and `RebaseDateTimeSuite`. Closes #28441 from MaxGekk/opt-rebase-common-threshold. Authored-by: Max Gekk <max.gekk@gmail.com> Signed-off-by: Wenchen Fan <wenchen@databricks.com>
2020-05-05 10:11:53 -04:00
date_trunc MM wholestage off 2605 2612 10 3.8 260.5 1.0X
date_trunc MM wholestage on 2501 2515 11 4.0 250.1 1.0X
[SPARK-29320][TESTS] Compare `sql/core` module in JDK8/11 (Part 1) ### What changes were proposed in this pull request? This PR regenerates the `sql/core` benchmarks in JDK8/11 to compare the result. In general, we compare the ratio instead of the time. However, in this PR, the average time is compared. This PR should be considered as a rough comparison. **A. EXPECTED CASES(JDK11 is faster in general)** - [x] BloomFilterBenchmark (JDK11 is faster except one case) - [x] BuiltInDataSourceWriteBenchmark (JDK11 is faster at CSV/ORC) - [x] CSVBenchmark (JDK11 is faster except five cases) - [x] ColumnarBatchBenchmark (JDK11 is faster at `boolean`/`string` and some cases in `int`/`array`) - [x] DatasetBenchmark (JDK11 is faster with `string`, but is slower for `long` type) - [x] ExternalAppendOnlyUnsafeRowArrayBenchmark (JDK11 is faster except two cases) - [x] ExtractBenchmark (JDK11 is faster except HOUR/MINUTE/SECOND/MILLISECONDS/MICROSECONDS) - [x] HashedRelationMetricsBenchmark (JDK11 is faster) - [x] JSONBenchmark (JDK11 is much faster except eight cases) - [x] JoinBenchmark (JDK11 is faster except five cases) - [x] OrcNestedSchemaPruningBenchmark (JDK11 is faster in nine cases) - [x] PrimitiveArrayBenchmark (JDK11 is faster) - [x] SortBenchmark (JDK11 is faster except `Arrays.sort` case) - [x] UDFBenchmark (N/A, values are too small) - [x] UnsafeArrayDataBenchmark (JDK11 is faster except one case) - [x] WideTableBenchmark (JDK11 is faster except two cases) **B. CASES WE NEED TO INVESTIGATE MORE LATER** - [x] AggregateBenchmark (JDK11 is slower in general) - [x] CompressionSchemeBenchmark (JDK11 is slower in general except `string`) - [x] DataSourceReadBenchmark (JDK11 is slower in general) - [x] DateTimeBenchmark (JDK11 is slightly slower in general except `parsing`) - [x] MakeDateTimeBenchmark (JDK11 is slower except two cases) - [x] MiscBenchmark (JDK11 is slower except ten cases) - [x] OrcV2NestedSchemaPruningBenchmark (JDK11 is slower) - [x] ParquetNestedSchemaPruningBenchmark (JDK11 is slower except six cases) - [x] RangeBenchmark (JDK11 is slower except one case) `FilterPushdownBenchmark/InExpressionBenchmark/WideSchemaBenchmark` will be compared later because it took long timer. ### Why are the changes needed? According to the result, there are some difference between JDK8/JDK11. This will be a baseline for the future improvement and comparison. Also, as a reproducible environment, the following environment is used. - Instance: `r3.xlarge` - OS: `CentOS Linux release 7.5.1804 (Core)` - JDK: - `OpenJDK Runtime Environment (build 1.8.0_222-b10)` - `OpenJDK Runtime Environment 18.9 (build 11.0.4+11-LTS)` ### Does this PR introduce any user-facing change? No. ### How was this patch tested? This is a test-only PR. We need to run benchmark. Closes #26003 from dongjoon-hyun/SPARK-29320. Authored-by: Dongjoon Hyun <dhyun@apple.com> Signed-off-by: Dongjoon Hyun <dhyun@apple.com>
2019-10-03 11:58:25 -04:00
[SPARK-31527][SQL][TESTS][FOLLOWUP] Fix the number of rows in `DateTimeBenchmark` ### What changes were proposed in this pull request? - Changed to the number of rows in benchmark cases from 3 to the actual number `N`. - Regenerated benchmark results in the environment: | Item | Description | | ---- | ----| | Region | us-west-2 (Oregon) | | Instance | r3.xlarge | | AMI | ubuntu/images/hvm-ssd/ubuntu-bionic-18.04-amd64-server-20190722.1 (ami-06f2f779464715dc5) | | Java | OpenJDK 64-Bit Server VM 1.8.0_242 and OpenJDK 64-Bit Server VM 11.0.6+10 | ### Why are the changes needed? The changes are needed to have: - Correct benchmark results - Base line for other perf improvements that can be checked in the same environment. ### Does this PR introduce _any_ user-facing change? No ### How was this patch tested? By running the benchmark and checking its output. Closes #28440 from MaxGekk/SPARK-31527-DateTimeBenchmark-followup. Authored-by: Max Gekk <max.gekk@gmail.com> Signed-off-by: Takeshi Yamamuro <yamamuro@apache.org>
2020-05-03 20:39:50 -04:00
OpenJDK 64-Bit Server VM 11.0.7+10-post-Ubuntu-2ubuntu218.04 on Linux 4.15.0-1063-aws
Intel(R) Xeon(R) CPU E5-2670 v2 @ 2.50GHz
[SPARK-29320][TESTS] Compare `sql/core` module in JDK8/11 (Part 1) ### What changes were proposed in this pull request? This PR regenerates the `sql/core` benchmarks in JDK8/11 to compare the result. In general, we compare the ratio instead of the time. However, in this PR, the average time is compared. This PR should be considered as a rough comparison. **A. EXPECTED CASES(JDK11 is faster in general)** - [x] BloomFilterBenchmark (JDK11 is faster except one case) - [x] BuiltInDataSourceWriteBenchmark (JDK11 is faster at CSV/ORC) - [x] CSVBenchmark (JDK11 is faster except five cases) - [x] ColumnarBatchBenchmark (JDK11 is faster at `boolean`/`string` and some cases in `int`/`array`) - [x] DatasetBenchmark (JDK11 is faster with `string`, but is slower for `long` type) - [x] ExternalAppendOnlyUnsafeRowArrayBenchmark (JDK11 is faster except two cases) - [x] ExtractBenchmark (JDK11 is faster except HOUR/MINUTE/SECOND/MILLISECONDS/MICROSECONDS) - [x] HashedRelationMetricsBenchmark (JDK11 is faster) - [x] JSONBenchmark (JDK11 is much faster except eight cases) - [x] JoinBenchmark (JDK11 is faster except five cases) - [x] OrcNestedSchemaPruningBenchmark (JDK11 is faster in nine cases) - [x] PrimitiveArrayBenchmark (JDK11 is faster) - [x] SortBenchmark (JDK11 is faster except `Arrays.sort` case) - [x] UDFBenchmark (N/A, values are too small) - [x] UnsafeArrayDataBenchmark (JDK11 is faster except one case) - [x] WideTableBenchmark (JDK11 is faster except two cases) **B. CASES WE NEED TO INVESTIGATE MORE LATER** - [x] AggregateBenchmark (JDK11 is slower in general) - [x] CompressionSchemeBenchmark (JDK11 is slower in general except `string`) - [x] DataSourceReadBenchmark (JDK11 is slower in general) - [x] DateTimeBenchmark (JDK11 is slightly slower in general except `parsing`) - [x] MakeDateTimeBenchmark (JDK11 is slower except two cases) - [x] MiscBenchmark (JDK11 is slower except ten cases) - [x] OrcV2NestedSchemaPruningBenchmark (JDK11 is slower) - [x] ParquetNestedSchemaPruningBenchmark (JDK11 is slower except six cases) - [x] RangeBenchmark (JDK11 is slower except one case) `FilterPushdownBenchmark/InExpressionBenchmark/WideSchemaBenchmark` will be compared later because it took long timer. ### Why are the changes needed? According to the result, there are some difference between JDK8/JDK11. This will be a baseline for the future improvement and comparison. Also, as a reproducible environment, the following environment is used. - Instance: `r3.xlarge` - OS: `CentOS Linux release 7.5.1804 (Core)` - JDK: - `OpenJDK Runtime Environment (build 1.8.0_222-b10)` - `OpenJDK Runtime Environment 18.9 (build 11.0.4+11-LTS)` ### Does this PR introduce any user-facing change? No. ### How was this patch tested? This is a test-only PR. We need to run benchmark. Closes #26003 from dongjoon-hyun/SPARK-29320. Authored-by: Dongjoon Hyun <dhyun@apple.com> Signed-off-by: Dongjoon Hyun <dhyun@apple.com>
2019-10-03 11:58:25 -04:00
date_trunc DAY: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative
------------------------------------------------------------------------------------------------------------------------
[SPARK-31630][SQL] Fix perf regression by skipping timestamps rebasing after some threshold ### What changes were proposed in this pull request? Skip timestamps rebasing after a global threshold when there is no difference between Julian and Gregorian calendars. This allows to avoid checking hash maps of switch points, and fixes perf regressions in `toJavaTimestamp()` and `fromJavaTimestamp()`. ### Why are the changes needed? The changes fix perf regressions of conversions to/from external type `java.sql.Timestamp`. Before (see the PR's results https://github.com/apache/spark/pull/28440): ``` ================================================================================================ Conversion from/to external types ================================================================================================ OpenJDK 64-Bit Server VM 1.8.0_252-8u252-b09-1~18.04-b09 on Linux 4.15.0-1063-aws Intel(R) Xeon(R) CPU E5-2670 v2 2.50GHz To/from Java's date-time: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative ------------------------------------------------------------------------------------------------------------------------ From java.sql.Timestamp 376 388 10 13.3 75.2 1.1X Collect java.sql.Timestamp 1878 1937 64 2.7 375.6 0.2X ``` After: ``` ================================================================================================ Conversion from/to external types ================================================================================================ OpenJDK 64-Bit Server VM 1.8.0_252-8u252-b09-1~18.04-b09 on Linux 4.15.0-1063-aws Intel(R) Xeon(R) CPU E5-2670 v2 2.50GHz To/from Java's date-time: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative ------------------------------------------------------------------------------------------------------------------------ From java.sql.Timestamp 249 264 24 20.1 49.8 1.7X Collect java.sql.Timestamp 1503 1523 24 3.3 300.5 0.3X ``` Perf improvements in average of: 1. From java.sql.Timestamp is ~ 34% 2. To java.sql.Timestamps is ~16% ### Does this PR introduce _any_ user-facing change? No ### How was this patch tested? By existing test suites `DateTimeUtilsSuite` and `RebaseDateTimeSuite`. Closes #28441 from MaxGekk/opt-rebase-common-threshold. Authored-by: Max Gekk <max.gekk@gmail.com> Signed-off-by: Wenchen Fan <wenchen@databricks.com>
2020-05-05 10:11:53 -04:00
date_trunc DAY wholestage off 2225 2229 5 4.5 222.5 1.0X
date_trunc DAY wholestage on 2184 2196 9 4.6 218.4 1.0X
[SPARK-29320][TESTS] Compare `sql/core` module in JDK8/11 (Part 1) ### What changes were proposed in this pull request? This PR regenerates the `sql/core` benchmarks in JDK8/11 to compare the result. In general, we compare the ratio instead of the time. However, in this PR, the average time is compared. This PR should be considered as a rough comparison. **A. EXPECTED CASES(JDK11 is faster in general)** - [x] BloomFilterBenchmark (JDK11 is faster except one case) - [x] BuiltInDataSourceWriteBenchmark (JDK11 is faster at CSV/ORC) - [x] CSVBenchmark (JDK11 is faster except five cases) - [x] ColumnarBatchBenchmark (JDK11 is faster at `boolean`/`string` and some cases in `int`/`array`) - [x] DatasetBenchmark (JDK11 is faster with `string`, but is slower for `long` type) - [x] ExternalAppendOnlyUnsafeRowArrayBenchmark (JDK11 is faster except two cases) - [x] ExtractBenchmark (JDK11 is faster except HOUR/MINUTE/SECOND/MILLISECONDS/MICROSECONDS) - [x] HashedRelationMetricsBenchmark (JDK11 is faster) - [x] JSONBenchmark (JDK11 is much faster except eight cases) - [x] JoinBenchmark (JDK11 is faster except five cases) - [x] OrcNestedSchemaPruningBenchmark (JDK11 is faster in nine cases) - [x] PrimitiveArrayBenchmark (JDK11 is faster) - [x] SortBenchmark (JDK11 is faster except `Arrays.sort` case) - [x] UDFBenchmark (N/A, values are too small) - [x] UnsafeArrayDataBenchmark (JDK11 is faster except one case) - [x] WideTableBenchmark (JDK11 is faster except two cases) **B. CASES WE NEED TO INVESTIGATE MORE LATER** - [x] AggregateBenchmark (JDK11 is slower in general) - [x] CompressionSchemeBenchmark (JDK11 is slower in general except `string`) - [x] DataSourceReadBenchmark (JDK11 is slower in general) - [x] DateTimeBenchmark (JDK11 is slightly slower in general except `parsing`) - [x] MakeDateTimeBenchmark (JDK11 is slower except two cases) - [x] MiscBenchmark (JDK11 is slower except ten cases) - [x] OrcV2NestedSchemaPruningBenchmark (JDK11 is slower) - [x] ParquetNestedSchemaPruningBenchmark (JDK11 is slower except six cases) - [x] RangeBenchmark (JDK11 is slower except one case) `FilterPushdownBenchmark/InExpressionBenchmark/WideSchemaBenchmark` will be compared later because it took long timer. ### Why are the changes needed? According to the result, there are some difference between JDK8/JDK11. This will be a baseline for the future improvement and comparison. Also, as a reproducible environment, the following environment is used. - Instance: `r3.xlarge` - OS: `CentOS Linux release 7.5.1804 (Core)` - JDK: - `OpenJDK Runtime Environment (build 1.8.0_222-b10)` - `OpenJDK Runtime Environment 18.9 (build 11.0.4+11-LTS)` ### Does this PR introduce any user-facing change? No. ### How was this patch tested? This is a test-only PR. We need to run benchmark. Closes #26003 from dongjoon-hyun/SPARK-29320. Authored-by: Dongjoon Hyun <dhyun@apple.com> Signed-off-by: Dongjoon Hyun <dhyun@apple.com>
2019-10-03 11:58:25 -04:00
[SPARK-31527][SQL][TESTS][FOLLOWUP] Fix the number of rows in `DateTimeBenchmark` ### What changes were proposed in this pull request? - Changed to the number of rows in benchmark cases from 3 to the actual number `N`. - Regenerated benchmark results in the environment: | Item | Description | | ---- | ----| | Region | us-west-2 (Oregon) | | Instance | r3.xlarge | | AMI | ubuntu/images/hvm-ssd/ubuntu-bionic-18.04-amd64-server-20190722.1 (ami-06f2f779464715dc5) | | Java | OpenJDK 64-Bit Server VM 1.8.0_242 and OpenJDK 64-Bit Server VM 11.0.6+10 | ### Why are the changes needed? The changes are needed to have: - Correct benchmark results - Base line for other perf improvements that can be checked in the same environment. ### Does this PR introduce _any_ user-facing change? No ### How was this patch tested? By running the benchmark and checking its output. Closes #28440 from MaxGekk/SPARK-31527-DateTimeBenchmark-followup. Authored-by: Max Gekk <max.gekk@gmail.com> Signed-off-by: Takeshi Yamamuro <yamamuro@apache.org>
2020-05-03 20:39:50 -04:00
OpenJDK 64-Bit Server VM 11.0.7+10-post-Ubuntu-2ubuntu218.04 on Linux 4.15.0-1063-aws
Intel(R) Xeon(R) CPU E5-2670 v2 @ 2.50GHz
[SPARK-29320][TESTS] Compare `sql/core` module in JDK8/11 (Part 1) ### What changes were proposed in this pull request? This PR regenerates the `sql/core` benchmarks in JDK8/11 to compare the result. In general, we compare the ratio instead of the time. However, in this PR, the average time is compared. This PR should be considered as a rough comparison. **A. EXPECTED CASES(JDK11 is faster in general)** - [x] BloomFilterBenchmark (JDK11 is faster except one case) - [x] BuiltInDataSourceWriteBenchmark (JDK11 is faster at CSV/ORC) - [x] CSVBenchmark (JDK11 is faster except five cases) - [x] ColumnarBatchBenchmark (JDK11 is faster at `boolean`/`string` and some cases in `int`/`array`) - [x] DatasetBenchmark (JDK11 is faster with `string`, but is slower for `long` type) - [x] ExternalAppendOnlyUnsafeRowArrayBenchmark (JDK11 is faster except two cases) - [x] ExtractBenchmark (JDK11 is faster except HOUR/MINUTE/SECOND/MILLISECONDS/MICROSECONDS) - [x] HashedRelationMetricsBenchmark (JDK11 is faster) - [x] JSONBenchmark (JDK11 is much faster except eight cases) - [x] JoinBenchmark (JDK11 is faster except five cases) - [x] OrcNestedSchemaPruningBenchmark (JDK11 is faster in nine cases) - [x] PrimitiveArrayBenchmark (JDK11 is faster) - [x] SortBenchmark (JDK11 is faster except `Arrays.sort` case) - [x] UDFBenchmark (N/A, values are too small) - [x] UnsafeArrayDataBenchmark (JDK11 is faster except one case) - [x] WideTableBenchmark (JDK11 is faster except two cases) **B. CASES WE NEED TO INVESTIGATE MORE LATER** - [x] AggregateBenchmark (JDK11 is slower in general) - [x] CompressionSchemeBenchmark (JDK11 is slower in general except `string`) - [x] DataSourceReadBenchmark (JDK11 is slower in general) - [x] DateTimeBenchmark (JDK11 is slightly slower in general except `parsing`) - [x] MakeDateTimeBenchmark (JDK11 is slower except two cases) - [x] MiscBenchmark (JDK11 is slower except ten cases) - [x] OrcV2NestedSchemaPruningBenchmark (JDK11 is slower) - [x] ParquetNestedSchemaPruningBenchmark (JDK11 is slower except six cases) - [x] RangeBenchmark (JDK11 is slower except one case) `FilterPushdownBenchmark/InExpressionBenchmark/WideSchemaBenchmark` will be compared later because it took long timer. ### Why are the changes needed? According to the result, there are some difference between JDK8/JDK11. This will be a baseline for the future improvement and comparison. Also, as a reproducible environment, the following environment is used. - Instance: `r3.xlarge` - OS: `CentOS Linux release 7.5.1804 (Core)` - JDK: - `OpenJDK Runtime Environment (build 1.8.0_222-b10)` - `OpenJDK Runtime Environment 18.9 (build 11.0.4+11-LTS)` ### Does this PR introduce any user-facing change? No. ### How was this patch tested? This is a test-only PR. We need to run benchmark. Closes #26003 from dongjoon-hyun/SPARK-29320. Authored-by: Dongjoon Hyun <dhyun@apple.com> Signed-off-by: Dongjoon Hyun <dhyun@apple.com>
2019-10-03 11:58:25 -04:00
date_trunc DD: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative
------------------------------------------------------------------------------------------------------------------------
[SPARK-31630][SQL] Fix perf regression by skipping timestamps rebasing after some threshold ### What changes were proposed in this pull request? Skip timestamps rebasing after a global threshold when there is no difference between Julian and Gregorian calendars. This allows to avoid checking hash maps of switch points, and fixes perf regressions in `toJavaTimestamp()` and `fromJavaTimestamp()`. ### Why are the changes needed? The changes fix perf regressions of conversions to/from external type `java.sql.Timestamp`. Before (see the PR's results https://github.com/apache/spark/pull/28440): ``` ================================================================================================ Conversion from/to external types ================================================================================================ OpenJDK 64-Bit Server VM 1.8.0_252-8u252-b09-1~18.04-b09 on Linux 4.15.0-1063-aws Intel(R) Xeon(R) CPU E5-2670 v2 2.50GHz To/from Java's date-time: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative ------------------------------------------------------------------------------------------------------------------------ From java.sql.Timestamp 376 388 10 13.3 75.2 1.1X Collect java.sql.Timestamp 1878 1937 64 2.7 375.6 0.2X ``` After: ``` ================================================================================================ Conversion from/to external types ================================================================================================ OpenJDK 64-Bit Server VM 1.8.0_252-8u252-b09-1~18.04-b09 on Linux 4.15.0-1063-aws Intel(R) Xeon(R) CPU E5-2670 v2 2.50GHz To/from Java's date-time: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative ------------------------------------------------------------------------------------------------------------------------ From java.sql.Timestamp 249 264 24 20.1 49.8 1.7X Collect java.sql.Timestamp 1503 1523 24 3.3 300.5 0.3X ``` Perf improvements in average of: 1. From java.sql.Timestamp is ~ 34% 2. To java.sql.Timestamps is ~16% ### Does this PR introduce _any_ user-facing change? No ### How was this patch tested? By existing test suites `DateTimeUtilsSuite` and `RebaseDateTimeSuite`. Closes #28441 from MaxGekk/opt-rebase-common-threshold. Authored-by: Max Gekk <max.gekk@gmail.com> Signed-off-by: Wenchen Fan <wenchen@databricks.com>
2020-05-05 10:11:53 -04:00
date_trunc DD wholestage off 2232 2236 6 4.5 223.2 1.0X
date_trunc DD wholestage on 2183 2190 6 4.6 218.3 1.0X
[SPARK-29320][TESTS] Compare `sql/core` module in JDK8/11 (Part 1) ### What changes were proposed in this pull request? This PR regenerates the `sql/core` benchmarks in JDK8/11 to compare the result. In general, we compare the ratio instead of the time. However, in this PR, the average time is compared. This PR should be considered as a rough comparison. **A. EXPECTED CASES(JDK11 is faster in general)** - [x] BloomFilterBenchmark (JDK11 is faster except one case) - [x] BuiltInDataSourceWriteBenchmark (JDK11 is faster at CSV/ORC) - [x] CSVBenchmark (JDK11 is faster except five cases) - [x] ColumnarBatchBenchmark (JDK11 is faster at `boolean`/`string` and some cases in `int`/`array`) - [x] DatasetBenchmark (JDK11 is faster with `string`, but is slower for `long` type) - [x] ExternalAppendOnlyUnsafeRowArrayBenchmark (JDK11 is faster except two cases) - [x] ExtractBenchmark (JDK11 is faster except HOUR/MINUTE/SECOND/MILLISECONDS/MICROSECONDS) - [x] HashedRelationMetricsBenchmark (JDK11 is faster) - [x] JSONBenchmark (JDK11 is much faster except eight cases) - [x] JoinBenchmark (JDK11 is faster except five cases) - [x] OrcNestedSchemaPruningBenchmark (JDK11 is faster in nine cases) - [x] PrimitiveArrayBenchmark (JDK11 is faster) - [x] SortBenchmark (JDK11 is faster except `Arrays.sort` case) - [x] UDFBenchmark (N/A, values are too small) - [x] UnsafeArrayDataBenchmark (JDK11 is faster except one case) - [x] WideTableBenchmark (JDK11 is faster except two cases) **B. CASES WE NEED TO INVESTIGATE MORE LATER** - [x] AggregateBenchmark (JDK11 is slower in general) - [x] CompressionSchemeBenchmark (JDK11 is slower in general except `string`) - [x] DataSourceReadBenchmark (JDK11 is slower in general) - [x] DateTimeBenchmark (JDK11 is slightly slower in general except `parsing`) - [x] MakeDateTimeBenchmark (JDK11 is slower except two cases) - [x] MiscBenchmark (JDK11 is slower except ten cases) - [x] OrcV2NestedSchemaPruningBenchmark (JDK11 is slower) - [x] ParquetNestedSchemaPruningBenchmark (JDK11 is slower except six cases) - [x] RangeBenchmark (JDK11 is slower except one case) `FilterPushdownBenchmark/InExpressionBenchmark/WideSchemaBenchmark` will be compared later because it took long timer. ### Why are the changes needed? According to the result, there are some difference between JDK8/JDK11. This will be a baseline for the future improvement and comparison. Also, as a reproducible environment, the following environment is used. - Instance: `r3.xlarge` - OS: `CentOS Linux release 7.5.1804 (Core)` - JDK: - `OpenJDK Runtime Environment (build 1.8.0_222-b10)` - `OpenJDK Runtime Environment 18.9 (build 11.0.4+11-LTS)` ### Does this PR introduce any user-facing change? No. ### How was this patch tested? This is a test-only PR. We need to run benchmark. Closes #26003 from dongjoon-hyun/SPARK-29320. Authored-by: Dongjoon Hyun <dhyun@apple.com> Signed-off-by: Dongjoon Hyun <dhyun@apple.com>
2019-10-03 11:58:25 -04:00
[SPARK-31527][SQL][TESTS][FOLLOWUP] Fix the number of rows in `DateTimeBenchmark` ### What changes were proposed in this pull request? - Changed to the number of rows in benchmark cases from 3 to the actual number `N`. - Regenerated benchmark results in the environment: | Item | Description | | ---- | ----| | Region | us-west-2 (Oregon) | | Instance | r3.xlarge | | AMI | ubuntu/images/hvm-ssd/ubuntu-bionic-18.04-amd64-server-20190722.1 (ami-06f2f779464715dc5) | | Java | OpenJDK 64-Bit Server VM 1.8.0_242 and OpenJDK 64-Bit Server VM 11.0.6+10 | ### Why are the changes needed? The changes are needed to have: - Correct benchmark results - Base line for other perf improvements that can be checked in the same environment. ### Does this PR introduce _any_ user-facing change? No ### How was this patch tested? By running the benchmark and checking its output. Closes #28440 from MaxGekk/SPARK-31527-DateTimeBenchmark-followup. Authored-by: Max Gekk <max.gekk@gmail.com> Signed-off-by: Takeshi Yamamuro <yamamuro@apache.org>
2020-05-03 20:39:50 -04:00
OpenJDK 64-Bit Server VM 11.0.7+10-post-Ubuntu-2ubuntu218.04 on Linux 4.15.0-1063-aws
Intel(R) Xeon(R) CPU E5-2670 v2 @ 2.50GHz
[SPARK-29320][TESTS] Compare `sql/core` module in JDK8/11 (Part 1) ### What changes were proposed in this pull request? This PR regenerates the `sql/core` benchmarks in JDK8/11 to compare the result. In general, we compare the ratio instead of the time. However, in this PR, the average time is compared. This PR should be considered as a rough comparison. **A. EXPECTED CASES(JDK11 is faster in general)** - [x] BloomFilterBenchmark (JDK11 is faster except one case) - [x] BuiltInDataSourceWriteBenchmark (JDK11 is faster at CSV/ORC) - [x] CSVBenchmark (JDK11 is faster except five cases) - [x] ColumnarBatchBenchmark (JDK11 is faster at `boolean`/`string` and some cases in `int`/`array`) - [x] DatasetBenchmark (JDK11 is faster with `string`, but is slower for `long` type) - [x] ExternalAppendOnlyUnsafeRowArrayBenchmark (JDK11 is faster except two cases) - [x] ExtractBenchmark (JDK11 is faster except HOUR/MINUTE/SECOND/MILLISECONDS/MICROSECONDS) - [x] HashedRelationMetricsBenchmark (JDK11 is faster) - [x] JSONBenchmark (JDK11 is much faster except eight cases) - [x] JoinBenchmark (JDK11 is faster except five cases) - [x] OrcNestedSchemaPruningBenchmark (JDK11 is faster in nine cases) - [x] PrimitiveArrayBenchmark (JDK11 is faster) - [x] SortBenchmark (JDK11 is faster except `Arrays.sort` case) - [x] UDFBenchmark (N/A, values are too small) - [x] UnsafeArrayDataBenchmark (JDK11 is faster except one case) - [x] WideTableBenchmark (JDK11 is faster except two cases) **B. CASES WE NEED TO INVESTIGATE MORE LATER** - [x] AggregateBenchmark (JDK11 is slower in general) - [x] CompressionSchemeBenchmark (JDK11 is slower in general except `string`) - [x] DataSourceReadBenchmark (JDK11 is slower in general) - [x] DateTimeBenchmark (JDK11 is slightly slower in general except `parsing`) - [x] MakeDateTimeBenchmark (JDK11 is slower except two cases) - [x] MiscBenchmark (JDK11 is slower except ten cases) - [x] OrcV2NestedSchemaPruningBenchmark (JDK11 is slower) - [x] ParquetNestedSchemaPruningBenchmark (JDK11 is slower except six cases) - [x] RangeBenchmark (JDK11 is slower except one case) `FilterPushdownBenchmark/InExpressionBenchmark/WideSchemaBenchmark` will be compared later because it took long timer. ### Why are the changes needed? According to the result, there are some difference between JDK8/JDK11. This will be a baseline for the future improvement and comparison. Also, as a reproducible environment, the following environment is used. - Instance: `r3.xlarge` - OS: `CentOS Linux release 7.5.1804 (Core)` - JDK: - `OpenJDK Runtime Environment (build 1.8.0_222-b10)` - `OpenJDK Runtime Environment 18.9 (build 11.0.4+11-LTS)` ### Does this PR introduce any user-facing change? No. ### How was this patch tested? This is a test-only PR. We need to run benchmark. Closes #26003 from dongjoon-hyun/SPARK-29320. Authored-by: Dongjoon Hyun <dhyun@apple.com> Signed-off-by: Dongjoon Hyun <dhyun@apple.com>
2019-10-03 11:58:25 -04:00
date_trunc HOUR: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative
------------------------------------------------------------------------------------------------------------------------
[SPARK-31630][SQL] Fix perf regression by skipping timestamps rebasing after some threshold ### What changes were proposed in this pull request? Skip timestamps rebasing after a global threshold when there is no difference between Julian and Gregorian calendars. This allows to avoid checking hash maps of switch points, and fixes perf regressions in `toJavaTimestamp()` and `fromJavaTimestamp()`. ### Why are the changes needed? The changes fix perf regressions of conversions to/from external type `java.sql.Timestamp`. Before (see the PR's results https://github.com/apache/spark/pull/28440): ``` ================================================================================================ Conversion from/to external types ================================================================================================ OpenJDK 64-Bit Server VM 1.8.0_252-8u252-b09-1~18.04-b09 on Linux 4.15.0-1063-aws Intel(R) Xeon(R) CPU E5-2670 v2 2.50GHz To/from Java's date-time: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative ------------------------------------------------------------------------------------------------------------------------ From java.sql.Timestamp 376 388 10 13.3 75.2 1.1X Collect java.sql.Timestamp 1878 1937 64 2.7 375.6 0.2X ``` After: ``` ================================================================================================ Conversion from/to external types ================================================================================================ OpenJDK 64-Bit Server VM 1.8.0_252-8u252-b09-1~18.04-b09 on Linux 4.15.0-1063-aws Intel(R) Xeon(R) CPU E5-2670 v2 2.50GHz To/from Java's date-time: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative ------------------------------------------------------------------------------------------------------------------------ From java.sql.Timestamp 249 264 24 20.1 49.8 1.7X Collect java.sql.Timestamp 1503 1523 24 3.3 300.5 0.3X ``` Perf improvements in average of: 1. From java.sql.Timestamp is ~ 34% 2. To java.sql.Timestamps is ~16% ### Does this PR introduce _any_ user-facing change? No ### How was this patch tested? By existing test suites `DateTimeUtilsSuite` and `RebaseDateTimeSuite`. Closes #28441 from MaxGekk/opt-rebase-common-threshold. Authored-by: Max Gekk <max.gekk@gmail.com> Signed-off-by: Wenchen Fan <wenchen@databricks.com>
2020-05-05 10:11:53 -04:00
date_trunc HOUR wholestage off 2194 2199 7 4.6 219.4 1.0X
date_trunc HOUR wholestage on 2160 2166 5 4.6 216.0 1.0X
[SPARK-29320][TESTS] Compare `sql/core` module in JDK8/11 (Part 1) ### What changes were proposed in this pull request? This PR regenerates the `sql/core` benchmarks in JDK8/11 to compare the result. In general, we compare the ratio instead of the time. However, in this PR, the average time is compared. This PR should be considered as a rough comparison. **A. EXPECTED CASES(JDK11 is faster in general)** - [x] BloomFilterBenchmark (JDK11 is faster except one case) - [x] BuiltInDataSourceWriteBenchmark (JDK11 is faster at CSV/ORC) - [x] CSVBenchmark (JDK11 is faster except five cases) - [x] ColumnarBatchBenchmark (JDK11 is faster at `boolean`/`string` and some cases in `int`/`array`) - [x] DatasetBenchmark (JDK11 is faster with `string`, but is slower for `long` type) - [x] ExternalAppendOnlyUnsafeRowArrayBenchmark (JDK11 is faster except two cases) - [x] ExtractBenchmark (JDK11 is faster except HOUR/MINUTE/SECOND/MILLISECONDS/MICROSECONDS) - [x] HashedRelationMetricsBenchmark (JDK11 is faster) - [x] JSONBenchmark (JDK11 is much faster except eight cases) - [x] JoinBenchmark (JDK11 is faster except five cases) - [x] OrcNestedSchemaPruningBenchmark (JDK11 is faster in nine cases) - [x] PrimitiveArrayBenchmark (JDK11 is faster) - [x] SortBenchmark (JDK11 is faster except `Arrays.sort` case) - [x] UDFBenchmark (N/A, values are too small) - [x] UnsafeArrayDataBenchmark (JDK11 is faster except one case) - [x] WideTableBenchmark (JDK11 is faster except two cases) **B. CASES WE NEED TO INVESTIGATE MORE LATER** - [x] AggregateBenchmark (JDK11 is slower in general) - [x] CompressionSchemeBenchmark (JDK11 is slower in general except `string`) - [x] DataSourceReadBenchmark (JDK11 is slower in general) - [x] DateTimeBenchmark (JDK11 is slightly slower in general except `parsing`) - [x] MakeDateTimeBenchmark (JDK11 is slower except two cases) - [x] MiscBenchmark (JDK11 is slower except ten cases) - [x] OrcV2NestedSchemaPruningBenchmark (JDK11 is slower) - [x] ParquetNestedSchemaPruningBenchmark (JDK11 is slower except six cases) - [x] RangeBenchmark (JDK11 is slower except one case) `FilterPushdownBenchmark/InExpressionBenchmark/WideSchemaBenchmark` will be compared later because it took long timer. ### Why are the changes needed? According to the result, there are some difference between JDK8/JDK11. This will be a baseline for the future improvement and comparison. Also, as a reproducible environment, the following environment is used. - Instance: `r3.xlarge` - OS: `CentOS Linux release 7.5.1804 (Core)` - JDK: - `OpenJDK Runtime Environment (build 1.8.0_222-b10)` - `OpenJDK Runtime Environment 18.9 (build 11.0.4+11-LTS)` ### Does this PR introduce any user-facing change? No. ### How was this patch tested? This is a test-only PR. We need to run benchmark. Closes #26003 from dongjoon-hyun/SPARK-29320. Authored-by: Dongjoon Hyun <dhyun@apple.com> Signed-off-by: Dongjoon Hyun <dhyun@apple.com>
2019-10-03 11:58:25 -04:00
[SPARK-31527][SQL][TESTS][FOLLOWUP] Fix the number of rows in `DateTimeBenchmark` ### What changes were proposed in this pull request? - Changed to the number of rows in benchmark cases from 3 to the actual number `N`. - Regenerated benchmark results in the environment: | Item | Description | | ---- | ----| | Region | us-west-2 (Oregon) | | Instance | r3.xlarge | | AMI | ubuntu/images/hvm-ssd/ubuntu-bionic-18.04-amd64-server-20190722.1 (ami-06f2f779464715dc5) | | Java | OpenJDK 64-Bit Server VM 1.8.0_242 and OpenJDK 64-Bit Server VM 11.0.6+10 | ### Why are the changes needed? The changes are needed to have: - Correct benchmark results - Base line for other perf improvements that can be checked in the same environment. ### Does this PR introduce _any_ user-facing change? No ### How was this patch tested? By running the benchmark and checking its output. Closes #28440 from MaxGekk/SPARK-31527-DateTimeBenchmark-followup. Authored-by: Max Gekk <max.gekk@gmail.com> Signed-off-by: Takeshi Yamamuro <yamamuro@apache.org>
2020-05-03 20:39:50 -04:00
OpenJDK 64-Bit Server VM 11.0.7+10-post-Ubuntu-2ubuntu218.04 on Linux 4.15.0-1063-aws
Intel(R) Xeon(R) CPU E5-2670 v2 @ 2.50GHz
[SPARK-29320][TESTS] Compare `sql/core` module in JDK8/11 (Part 1) ### What changes were proposed in this pull request? This PR regenerates the `sql/core` benchmarks in JDK8/11 to compare the result. In general, we compare the ratio instead of the time. However, in this PR, the average time is compared. This PR should be considered as a rough comparison. **A. EXPECTED CASES(JDK11 is faster in general)** - [x] BloomFilterBenchmark (JDK11 is faster except one case) - [x] BuiltInDataSourceWriteBenchmark (JDK11 is faster at CSV/ORC) - [x] CSVBenchmark (JDK11 is faster except five cases) - [x] ColumnarBatchBenchmark (JDK11 is faster at `boolean`/`string` and some cases in `int`/`array`) - [x] DatasetBenchmark (JDK11 is faster with `string`, but is slower for `long` type) - [x] ExternalAppendOnlyUnsafeRowArrayBenchmark (JDK11 is faster except two cases) - [x] ExtractBenchmark (JDK11 is faster except HOUR/MINUTE/SECOND/MILLISECONDS/MICROSECONDS) - [x] HashedRelationMetricsBenchmark (JDK11 is faster) - [x] JSONBenchmark (JDK11 is much faster except eight cases) - [x] JoinBenchmark (JDK11 is faster except five cases) - [x] OrcNestedSchemaPruningBenchmark (JDK11 is faster in nine cases) - [x] PrimitiveArrayBenchmark (JDK11 is faster) - [x] SortBenchmark (JDK11 is faster except `Arrays.sort` case) - [x] UDFBenchmark (N/A, values are too small) - [x] UnsafeArrayDataBenchmark (JDK11 is faster except one case) - [x] WideTableBenchmark (JDK11 is faster except two cases) **B. CASES WE NEED TO INVESTIGATE MORE LATER** - [x] AggregateBenchmark (JDK11 is slower in general) - [x] CompressionSchemeBenchmark (JDK11 is slower in general except `string`) - [x] DataSourceReadBenchmark (JDK11 is slower in general) - [x] DateTimeBenchmark (JDK11 is slightly slower in general except `parsing`) - [x] MakeDateTimeBenchmark (JDK11 is slower except two cases) - [x] MiscBenchmark (JDK11 is slower except ten cases) - [x] OrcV2NestedSchemaPruningBenchmark (JDK11 is slower) - [x] ParquetNestedSchemaPruningBenchmark (JDK11 is slower except six cases) - [x] RangeBenchmark (JDK11 is slower except one case) `FilterPushdownBenchmark/InExpressionBenchmark/WideSchemaBenchmark` will be compared later because it took long timer. ### Why are the changes needed? According to the result, there are some difference between JDK8/JDK11. This will be a baseline for the future improvement and comparison. Also, as a reproducible environment, the following environment is used. - Instance: `r3.xlarge` - OS: `CentOS Linux release 7.5.1804 (Core)` - JDK: - `OpenJDK Runtime Environment (build 1.8.0_222-b10)` - `OpenJDK Runtime Environment 18.9 (build 11.0.4+11-LTS)` ### Does this PR introduce any user-facing change? No. ### How was this patch tested? This is a test-only PR. We need to run benchmark. Closes #26003 from dongjoon-hyun/SPARK-29320. Authored-by: Dongjoon Hyun <dhyun@apple.com> Signed-off-by: Dongjoon Hyun <dhyun@apple.com>
2019-10-03 11:58:25 -04:00
date_trunc MINUTE: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative
------------------------------------------------------------------------------------------------------------------------
[SPARK-31630][SQL] Fix perf regression by skipping timestamps rebasing after some threshold ### What changes were proposed in this pull request? Skip timestamps rebasing after a global threshold when there is no difference between Julian and Gregorian calendars. This allows to avoid checking hash maps of switch points, and fixes perf regressions in `toJavaTimestamp()` and `fromJavaTimestamp()`. ### Why are the changes needed? The changes fix perf regressions of conversions to/from external type `java.sql.Timestamp`. Before (see the PR's results https://github.com/apache/spark/pull/28440): ``` ================================================================================================ Conversion from/to external types ================================================================================================ OpenJDK 64-Bit Server VM 1.8.0_252-8u252-b09-1~18.04-b09 on Linux 4.15.0-1063-aws Intel(R) Xeon(R) CPU E5-2670 v2 2.50GHz To/from Java's date-time: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative ------------------------------------------------------------------------------------------------------------------------ From java.sql.Timestamp 376 388 10 13.3 75.2 1.1X Collect java.sql.Timestamp 1878 1937 64 2.7 375.6 0.2X ``` After: ``` ================================================================================================ Conversion from/to external types ================================================================================================ OpenJDK 64-Bit Server VM 1.8.0_252-8u252-b09-1~18.04-b09 on Linux 4.15.0-1063-aws Intel(R) Xeon(R) CPU E5-2670 v2 2.50GHz To/from Java's date-time: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative ------------------------------------------------------------------------------------------------------------------------ From java.sql.Timestamp 249 264 24 20.1 49.8 1.7X Collect java.sql.Timestamp 1503 1523 24 3.3 300.5 0.3X ``` Perf improvements in average of: 1. From java.sql.Timestamp is ~ 34% 2. To java.sql.Timestamps is ~16% ### Does this PR introduce _any_ user-facing change? No ### How was this patch tested? By existing test suites `DateTimeUtilsSuite` and `RebaseDateTimeSuite`. Closes #28441 from MaxGekk/opt-rebase-common-threshold. Authored-by: Max Gekk <max.gekk@gmail.com> Signed-off-by: Wenchen Fan <wenchen@databricks.com>
2020-05-05 10:11:53 -04:00
date_trunc MINUTE wholestage off 390 396 9 25.7 39.0 1.0X
date_trunc MINUTE wholestage on 331 337 7 30.2 33.1 1.2X
[SPARK-29320][TESTS] Compare `sql/core` module in JDK8/11 (Part 1) ### What changes were proposed in this pull request? This PR regenerates the `sql/core` benchmarks in JDK8/11 to compare the result. In general, we compare the ratio instead of the time. However, in this PR, the average time is compared. This PR should be considered as a rough comparison. **A. EXPECTED CASES(JDK11 is faster in general)** - [x] BloomFilterBenchmark (JDK11 is faster except one case) - [x] BuiltInDataSourceWriteBenchmark (JDK11 is faster at CSV/ORC) - [x] CSVBenchmark (JDK11 is faster except five cases) - [x] ColumnarBatchBenchmark (JDK11 is faster at `boolean`/`string` and some cases in `int`/`array`) - [x] DatasetBenchmark (JDK11 is faster with `string`, but is slower for `long` type) - [x] ExternalAppendOnlyUnsafeRowArrayBenchmark (JDK11 is faster except two cases) - [x] ExtractBenchmark (JDK11 is faster except HOUR/MINUTE/SECOND/MILLISECONDS/MICROSECONDS) - [x] HashedRelationMetricsBenchmark (JDK11 is faster) - [x] JSONBenchmark (JDK11 is much faster except eight cases) - [x] JoinBenchmark (JDK11 is faster except five cases) - [x] OrcNestedSchemaPruningBenchmark (JDK11 is faster in nine cases) - [x] PrimitiveArrayBenchmark (JDK11 is faster) - [x] SortBenchmark (JDK11 is faster except `Arrays.sort` case) - [x] UDFBenchmark (N/A, values are too small) - [x] UnsafeArrayDataBenchmark (JDK11 is faster except one case) - [x] WideTableBenchmark (JDK11 is faster except two cases) **B. CASES WE NEED TO INVESTIGATE MORE LATER** - [x] AggregateBenchmark (JDK11 is slower in general) - [x] CompressionSchemeBenchmark (JDK11 is slower in general except `string`) - [x] DataSourceReadBenchmark (JDK11 is slower in general) - [x] DateTimeBenchmark (JDK11 is slightly slower in general except `parsing`) - [x] MakeDateTimeBenchmark (JDK11 is slower except two cases) - [x] MiscBenchmark (JDK11 is slower except ten cases) - [x] OrcV2NestedSchemaPruningBenchmark (JDK11 is slower) - [x] ParquetNestedSchemaPruningBenchmark (JDK11 is slower except six cases) - [x] RangeBenchmark (JDK11 is slower except one case) `FilterPushdownBenchmark/InExpressionBenchmark/WideSchemaBenchmark` will be compared later because it took long timer. ### Why are the changes needed? According to the result, there are some difference between JDK8/JDK11. This will be a baseline for the future improvement and comparison. Also, as a reproducible environment, the following environment is used. - Instance: `r3.xlarge` - OS: `CentOS Linux release 7.5.1804 (Core)` - JDK: - `OpenJDK Runtime Environment (build 1.8.0_222-b10)` - `OpenJDK Runtime Environment 18.9 (build 11.0.4+11-LTS)` ### Does this PR introduce any user-facing change? No. ### How was this patch tested? This is a test-only PR. We need to run benchmark. Closes #26003 from dongjoon-hyun/SPARK-29320. Authored-by: Dongjoon Hyun <dhyun@apple.com> Signed-off-by: Dongjoon Hyun <dhyun@apple.com>
2019-10-03 11:58:25 -04:00
[SPARK-31527][SQL][TESTS][FOLLOWUP] Fix the number of rows in `DateTimeBenchmark` ### What changes were proposed in this pull request? - Changed to the number of rows in benchmark cases from 3 to the actual number `N`. - Regenerated benchmark results in the environment: | Item | Description | | ---- | ----| | Region | us-west-2 (Oregon) | | Instance | r3.xlarge | | AMI | ubuntu/images/hvm-ssd/ubuntu-bionic-18.04-amd64-server-20190722.1 (ami-06f2f779464715dc5) | | Java | OpenJDK 64-Bit Server VM 1.8.0_242 and OpenJDK 64-Bit Server VM 11.0.6+10 | ### Why are the changes needed? The changes are needed to have: - Correct benchmark results - Base line for other perf improvements that can be checked in the same environment. ### Does this PR introduce _any_ user-facing change? No ### How was this patch tested? By running the benchmark and checking its output. Closes #28440 from MaxGekk/SPARK-31527-DateTimeBenchmark-followup. Authored-by: Max Gekk <max.gekk@gmail.com> Signed-off-by: Takeshi Yamamuro <yamamuro@apache.org>
2020-05-03 20:39:50 -04:00
OpenJDK 64-Bit Server VM 11.0.7+10-post-Ubuntu-2ubuntu218.04 on Linux 4.15.0-1063-aws
Intel(R) Xeon(R) CPU E5-2670 v2 @ 2.50GHz
[SPARK-29320][TESTS] Compare `sql/core` module in JDK8/11 (Part 1) ### What changes were proposed in this pull request? This PR regenerates the `sql/core` benchmarks in JDK8/11 to compare the result. In general, we compare the ratio instead of the time. However, in this PR, the average time is compared. This PR should be considered as a rough comparison. **A. EXPECTED CASES(JDK11 is faster in general)** - [x] BloomFilterBenchmark (JDK11 is faster except one case) - [x] BuiltInDataSourceWriteBenchmark (JDK11 is faster at CSV/ORC) - [x] CSVBenchmark (JDK11 is faster except five cases) - [x] ColumnarBatchBenchmark (JDK11 is faster at `boolean`/`string` and some cases in `int`/`array`) - [x] DatasetBenchmark (JDK11 is faster with `string`, but is slower for `long` type) - [x] ExternalAppendOnlyUnsafeRowArrayBenchmark (JDK11 is faster except two cases) - [x] ExtractBenchmark (JDK11 is faster except HOUR/MINUTE/SECOND/MILLISECONDS/MICROSECONDS) - [x] HashedRelationMetricsBenchmark (JDK11 is faster) - [x] JSONBenchmark (JDK11 is much faster except eight cases) - [x] JoinBenchmark (JDK11 is faster except five cases) - [x] OrcNestedSchemaPruningBenchmark (JDK11 is faster in nine cases) - [x] PrimitiveArrayBenchmark (JDK11 is faster) - [x] SortBenchmark (JDK11 is faster except `Arrays.sort` case) - [x] UDFBenchmark (N/A, values are too small) - [x] UnsafeArrayDataBenchmark (JDK11 is faster except one case) - [x] WideTableBenchmark (JDK11 is faster except two cases) **B. CASES WE NEED TO INVESTIGATE MORE LATER** - [x] AggregateBenchmark (JDK11 is slower in general) - [x] CompressionSchemeBenchmark (JDK11 is slower in general except `string`) - [x] DataSourceReadBenchmark (JDK11 is slower in general) - [x] DateTimeBenchmark (JDK11 is slightly slower in general except `parsing`) - [x] MakeDateTimeBenchmark (JDK11 is slower except two cases) - [x] MiscBenchmark (JDK11 is slower except ten cases) - [x] OrcV2NestedSchemaPruningBenchmark (JDK11 is slower) - [x] ParquetNestedSchemaPruningBenchmark (JDK11 is slower except six cases) - [x] RangeBenchmark (JDK11 is slower except one case) `FilterPushdownBenchmark/InExpressionBenchmark/WideSchemaBenchmark` will be compared later because it took long timer. ### Why are the changes needed? According to the result, there are some difference between JDK8/JDK11. This will be a baseline for the future improvement and comparison. Also, as a reproducible environment, the following environment is used. - Instance: `r3.xlarge` - OS: `CentOS Linux release 7.5.1804 (Core)` - JDK: - `OpenJDK Runtime Environment (build 1.8.0_222-b10)` - `OpenJDK Runtime Environment 18.9 (build 11.0.4+11-LTS)` ### Does this PR introduce any user-facing change? No. ### How was this patch tested? This is a test-only PR. We need to run benchmark. Closes #26003 from dongjoon-hyun/SPARK-29320. Authored-by: Dongjoon Hyun <dhyun@apple.com> Signed-off-by: Dongjoon Hyun <dhyun@apple.com>
2019-10-03 11:58:25 -04:00
date_trunc SECOND: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative
------------------------------------------------------------------------------------------------------------------------
[SPARK-31630][SQL] Fix perf regression by skipping timestamps rebasing after some threshold ### What changes were proposed in this pull request? Skip timestamps rebasing after a global threshold when there is no difference between Julian and Gregorian calendars. This allows to avoid checking hash maps of switch points, and fixes perf regressions in `toJavaTimestamp()` and `fromJavaTimestamp()`. ### Why are the changes needed? The changes fix perf regressions of conversions to/from external type `java.sql.Timestamp`. Before (see the PR's results https://github.com/apache/spark/pull/28440): ``` ================================================================================================ Conversion from/to external types ================================================================================================ OpenJDK 64-Bit Server VM 1.8.0_252-8u252-b09-1~18.04-b09 on Linux 4.15.0-1063-aws Intel(R) Xeon(R) CPU E5-2670 v2 2.50GHz To/from Java's date-time: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative ------------------------------------------------------------------------------------------------------------------------ From java.sql.Timestamp 376 388 10 13.3 75.2 1.1X Collect java.sql.Timestamp 1878 1937 64 2.7 375.6 0.2X ``` After: ``` ================================================================================================ Conversion from/to external types ================================================================================================ OpenJDK 64-Bit Server VM 1.8.0_252-8u252-b09-1~18.04-b09 on Linux 4.15.0-1063-aws Intel(R) Xeon(R) CPU E5-2670 v2 2.50GHz To/from Java's date-time: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative ------------------------------------------------------------------------------------------------------------------------ From java.sql.Timestamp 249 264 24 20.1 49.8 1.7X Collect java.sql.Timestamp 1503 1523 24 3.3 300.5 0.3X ``` Perf improvements in average of: 1. From java.sql.Timestamp is ~ 34% 2. To java.sql.Timestamps is ~16% ### Does this PR introduce _any_ user-facing change? No ### How was this patch tested? By existing test suites `DateTimeUtilsSuite` and `RebaseDateTimeSuite`. Closes #28441 from MaxGekk/opt-rebase-common-threshold. Authored-by: Max Gekk <max.gekk@gmail.com> Signed-off-by: Wenchen Fan <wenchen@databricks.com>
2020-05-05 10:11:53 -04:00
date_trunc SECOND wholestage off 375 381 8 26.7 37.5 1.0X
date_trunc SECOND wholestage on 332 346 14 30.1 33.2 1.1X
[SPARK-29320][TESTS] Compare `sql/core` module in JDK8/11 (Part 1) ### What changes were proposed in this pull request? This PR regenerates the `sql/core` benchmarks in JDK8/11 to compare the result. In general, we compare the ratio instead of the time. However, in this PR, the average time is compared. This PR should be considered as a rough comparison. **A. EXPECTED CASES(JDK11 is faster in general)** - [x] BloomFilterBenchmark (JDK11 is faster except one case) - [x] BuiltInDataSourceWriteBenchmark (JDK11 is faster at CSV/ORC) - [x] CSVBenchmark (JDK11 is faster except five cases) - [x] ColumnarBatchBenchmark (JDK11 is faster at `boolean`/`string` and some cases in `int`/`array`) - [x] DatasetBenchmark (JDK11 is faster with `string`, but is slower for `long` type) - [x] ExternalAppendOnlyUnsafeRowArrayBenchmark (JDK11 is faster except two cases) - [x] ExtractBenchmark (JDK11 is faster except HOUR/MINUTE/SECOND/MILLISECONDS/MICROSECONDS) - [x] HashedRelationMetricsBenchmark (JDK11 is faster) - [x] JSONBenchmark (JDK11 is much faster except eight cases) - [x] JoinBenchmark (JDK11 is faster except five cases) - [x] OrcNestedSchemaPruningBenchmark (JDK11 is faster in nine cases) - [x] PrimitiveArrayBenchmark (JDK11 is faster) - [x] SortBenchmark (JDK11 is faster except `Arrays.sort` case) - [x] UDFBenchmark (N/A, values are too small) - [x] UnsafeArrayDataBenchmark (JDK11 is faster except one case) - [x] WideTableBenchmark (JDK11 is faster except two cases) **B. CASES WE NEED TO INVESTIGATE MORE LATER** - [x] AggregateBenchmark (JDK11 is slower in general) - [x] CompressionSchemeBenchmark (JDK11 is slower in general except `string`) - [x] DataSourceReadBenchmark (JDK11 is slower in general) - [x] DateTimeBenchmark (JDK11 is slightly slower in general except `parsing`) - [x] MakeDateTimeBenchmark (JDK11 is slower except two cases) - [x] MiscBenchmark (JDK11 is slower except ten cases) - [x] OrcV2NestedSchemaPruningBenchmark (JDK11 is slower) - [x] ParquetNestedSchemaPruningBenchmark (JDK11 is slower except six cases) - [x] RangeBenchmark (JDK11 is slower except one case) `FilterPushdownBenchmark/InExpressionBenchmark/WideSchemaBenchmark` will be compared later because it took long timer. ### Why are the changes needed? According to the result, there are some difference between JDK8/JDK11. This will be a baseline for the future improvement and comparison. Also, as a reproducible environment, the following environment is used. - Instance: `r3.xlarge` - OS: `CentOS Linux release 7.5.1804 (Core)` - JDK: - `OpenJDK Runtime Environment (build 1.8.0_222-b10)` - `OpenJDK Runtime Environment 18.9 (build 11.0.4+11-LTS)` ### Does this PR introduce any user-facing change? No. ### How was this patch tested? This is a test-only PR. We need to run benchmark. Closes #26003 from dongjoon-hyun/SPARK-29320. Authored-by: Dongjoon Hyun <dhyun@apple.com> Signed-off-by: Dongjoon Hyun <dhyun@apple.com>
2019-10-03 11:58:25 -04:00
[SPARK-31527][SQL][TESTS][FOLLOWUP] Fix the number of rows in `DateTimeBenchmark` ### What changes were proposed in this pull request? - Changed to the number of rows in benchmark cases from 3 to the actual number `N`. - Regenerated benchmark results in the environment: | Item | Description | | ---- | ----| | Region | us-west-2 (Oregon) | | Instance | r3.xlarge | | AMI | ubuntu/images/hvm-ssd/ubuntu-bionic-18.04-amd64-server-20190722.1 (ami-06f2f779464715dc5) | | Java | OpenJDK 64-Bit Server VM 1.8.0_242 and OpenJDK 64-Bit Server VM 11.0.6+10 | ### Why are the changes needed? The changes are needed to have: - Correct benchmark results - Base line for other perf improvements that can be checked in the same environment. ### Does this PR introduce _any_ user-facing change? No ### How was this patch tested? By running the benchmark and checking its output. Closes #28440 from MaxGekk/SPARK-31527-DateTimeBenchmark-followup. Authored-by: Max Gekk <max.gekk@gmail.com> Signed-off-by: Takeshi Yamamuro <yamamuro@apache.org>
2020-05-03 20:39:50 -04:00
OpenJDK 64-Bit Server VM 11.0.7+10-post-Ubuntu-2ubuntu218.04 on Linux 4.15.0-1063-aws
Intel(R) Xeon(R) CPU E5-2670 v2 @ 2.50GHz
[SPARK-29320][TESTS] Compare `sql/core` module in JDK8/11 (Part 1) ### What changes were proposed in this pull request? This PR regenerates the `sql/core` benchmarks in JDK8/11 to compare the result. In general, we compare the ratio instead of the time. However, in this PR, the average time is compared. This PR should be considered as a rough comparison. **A. EXPECTED CASES(JDK11 is faster in general)** - [x] BloomFilterBenchmark (JDK11 is faster except one case) - [x] BuiltInDataSourceWriteBenchmark (JDK11 is faster at CSV/ORC) - [x] CSVBenchmark (JDK11 is faster except five cases) - [x] ColumnarBatchBenchmark (JDK11 is faster at `boolean`/`string` and some cases in `int`/`array`) - [x] DatasetBenchmark (JDK11 is faster with `string`, but is slower for `long` type) - [x] ExternalAppendOnlyUnsafeRowArrayBenchmark (JDK11 is faster except two cases) - [x] ExtractBenchmark (JDK11 is faster except HOUR/MINUTE/SECOND/MILLISECONDS/MICROSECONDS) - [x] HashedRelationMetricsBenchmark (JDK11 is faster) - [x] JSONBenchmark (JDK11 is much faster except eight cases) - [x] JoinBenchmark (JDK11 is faster except five cases) - [x] OrcNestedSchemaPruningBenchmark (JDK11 is faster in nine cases) - [x] PrimitiveArrayBenchmark (JDK11 is faster) - [x] SortBenchmark (JDK11 is faster except `Arrays.sort` case) - [x] UDFBenchmark (N/A, values are too small) - [x] UnsafeArrayDataBenchmark (JDK11 is faster except one case) - [x] WideTableBenchmark (JDK11 is faster except two cases) **B. CASES WE NEED TO INVESTIGATE MORE LATER** - [x] AggregateBenchmark (JDK11 is slower in general) - [x] CompressionSchemeBenchmark (JDK11 is slower in general except `string`) - [x] DataSourceReadBenchmark (JDK11 is slower in general) - [x] DateTimeBenchmark (JDK11 is slightly slower in general except `parsing`) - [x] MakeDateTimeBenchmark (JDK11 is slower except two cases) - [x] MiscBenchmark (JDK11 is slower except ten cases) - [x] OrcV2NestedSchemaPruningBenchmark (JDK11 is slower) - [x] ParquetNestedSchemaPruningBenchmark (JDK11 is slower except six cases) - [x] RangeBenchmark (JDK11 is slower except one case) `FilterPushdownBenchmark/InExpressionBenchmark/WideSchemaBenchmark` will be compared later because it took long timer. ### Why are the changes needed? According to the result, there are some difference between JDK8/JDK11. This will be a baseline for the future improvement and comparison. Also, as a reproducible environment, the following environment is used. - Instance: `r3.xlarge` - OS: `CentOS Linux release 7.5.1804 (Core)` - JDK: - `OpenJDK Runtime Environment (build 1.8.0_222-b10)` - `OpenJDK Runtime Environment 18.9 (build 11.0.4+11-LTS)` ### Does this PR introduce any user-facing change? No. ### How was this patch tested? This is a test-only PR. We need to run benchmark. Closes #26003 from dongjoon-hyun/SPARK-29320. Authored-by: Dongjoon Hyun <dhyun@apple.com> Signed-off-by: Dongjoon Hyun <dhyun@apple.com>
2019-10-03 11:58:25 -04:00
date_trunc WEEK: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative
------------------------------------------------------------------------------------------------------------------------
[SPARK-31630][SQL] Fix perf regression by skipping timestamps rebasing after some threshold ### What changes were proposed in this pull request? Skip timestamps rebasing after a global threshold when there is no difference between Julian and Gregorian calendars. This allows to avoid checking hash maps of switch points, and fixes perf regressions in `toJavaTimestamp()` and `fromJavaTimestamp()`. ### Why are the changes needed? The changes fix perf regressions of conversions to/from external type `java.sql.Timestamp`. Before (see the PR's results https://github.com/apache/spark/pull/28440): ``` ================================================================================================ Conversion from/to external types ================================================================================================ OpenJDK 64-Bit Server VM 1.8.0_252-8u252-b09-1~18.04-b09 on Linux 4.15.0-1063-aws Intel(R) Xeon(R) CPU E5-2670 v2 2.50GHz To/from Java's date-time: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative ------------------------------------------------------------------------------------------------------------------------ From java.sql.Timestamp 376 388 10 13.3 75.2 1.1X Collect java.sql.Timestamp 1878 1937 64 2.7 375.6 0.2X ``` After: ``` ================================================================================================ Conversion from/to external types ================================================================================================ OpenJDK 64-Bit Server VM 1.8.0_252-8u252-b09-1~18.04-b09 on Linux 4.15.0-1063-aws Intel(R) Xeon(R) CPU E5-2670 v2 2.50GHz To/from Java's date-time: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative ------------------------------------------------------------------------------------------------------------------------ From java.sql.Timestamp 249 264 24 20.1 49.8 1.7X Collect java.sql.Timestamp 1503 1523 24 3.3 300.5 0.3X ``` Perf improvements in average of: 1. From java.sql.Timestamp is ~ 34% 2. To java.sql.Timestamps is ~16% ### Does this PR introduce _any_ user-facing change? No ### How was this patch tested? By existing test suites `DateTimeUtilsSuite` and `RebaseDateTimeSuite`. Closes #28441 from MaxGekk/opt-rebase-common-threshold. Authored-by: Max Gekk <max.gekk@gmail.com> Signed-off-by: Wenchen Fan <wenchen@databricks.com>
2020-05-05 10:11:53 -04:00
date_trunc WEEK wholestage off 2439 2443 6 4.1 243.9 1.0X
date_trunc WEEK wholestage on 2390 2409 32 4.2 239.0 1.0X
[SPARK-29320][TESTS] Compare `sql/core` module in JDK8/11 (Part 1) ### What changes were proposed in this pull request? This PR regenerates the `sql/core` benchmarks in JDK8/11 to compare the result. In general, we compare the ratio instead of the time. However, in this PR, the average time is compared. This PR should be considered as a rough comparison. **A. EXPECTED CASES(JDK11 is faster in general)** - [x] BloomFilterBenchmark (JDK11 is faster except one case) - [x] BuiltInDataSourceWriteBenchmark (JDK11 is faster at CSV/ORC) - [x] CSVBenchmark (JDK11 is faster except five cases) - [x] ColumnarBatchBenchmark (JDK11 is faster at `boolean`/`string` and some cases in `int`/`array`) - [x] DatasetBenchmark (JDK11 is faster with `string`, but is slower for `long` type) - [x] ExternalAppendOnlyUnsafeRowArrayBenchmark (JDK11 is faster except two cases) - [x] ExtractBenchmark (JDK11 is faster except HOUR/MINUTE/SECOND/MILLISECONDS/MICROSECONDS) - [x] HashedRelationMetricsBenchmark (JDK11 is faster) - [x] JSONBenchmark (JDK11 is much faster except eight cases) - [x] JoinBenchmark (JDK11 is faster except five cases) - [x] OrcNestedSchemaPruningBenchmark (JDK11 is faster in nine cases) - [x] PrimitiveArrayBenchmark (JDK11 is faster) - [x] SortBenchmark (JDK11 is faster except `Arrays.sort` case) - [x] UDFBenchmark (N/A, values are too small) - [x] UnsafeArrayDataBenchmark (JDK11 is faster except one case) - [x] WideTableBenchmark (JDK11 is faster except two cases) **B. CASES WE NEED TO INVESTIGATE MORE LATER** - [x] AggregateBenchmark (JDK11 is slower in general) - [x] CompressionSchemeBenchmark (JDK11 is slower in general except `string`) - [x] DataSourceReadBenchmark (JDK11 is slower in general) - [x] DateTimeBenchmark (JDK11 is slightly slower in general except `parsing`) - [x] MakeDateTimeBenchmark (JDK11 is slower except two cases) - [x] MiscBenchmark (JDK11 is slower except ten cases) - [x] OrcV2NestedSchemaPruningBenchmark (JDK11 is slower) - [x] ParquetNestedSchemaPruningBenchmark (JDK11 is slower except six cases) - [x] RangeBenchmark (JDK11 is slower except one case) `FilterPushdownBenchmark/InExpressionBenchmark/WideSchemaBenchmark` will be compared later because it took long timer. ### Why are the changes needed? According to the result, there are some difference between JDK8/JDK11. This will be a baseline for the future improvement and comparison. Also, as a reproducible environment, the following environment is used. - Instance: `r3.xlarge` - OS: `CentOS Linux release 7.5.1804 (Core)` - JDK: - `OpenJDK Runtime Environment (build 1.8.0_222-b10)` - `OpenJDK Runtime Environment 18.9 (build 11.0.4+11-LTS)` ### Does this PR introduce any user-facing change? No. ### How was this patch tested? This is a test-only PR. We need to run benchmark. Closes #26003 from dongjoon-hyun/SPARK-29320. Authored-by: Dongjoon Hyun <dhyun@apple.com> Signed-off-by: Dongjoon Hyun <dhyun@apple.com>
2019-10-03 11:58:25 -04:00
[SPARK-31527][SQL][TESTS][FOLLOWUP] Fix the number of rows in `DateTimeBenchmark` ### What changes were proposed in this pull request? - Changed to the number of rows in benchmark cases from 3 to the actual number `N`. - Regenerated benchmark results in the environment: | Item | Description | | ---- | ----| | Region | us-west-2 (Oregon) | | Instance | r3.xlarge | | AMI | ubuntu/images/hvm-ssd/ubuntu-bionic-18.04-amd64-server-20190722.1 (ami-06f2f779464715dc5) | | Java | OpenJDK 64-Bit Server VM 1.8.0_242 and OpenJDK 64-Bit Server VM 11.0.6+10 | ### Why are the changes needed? The changes are needed to have: - Correct benchmark results - Base line for other perf improvements that can be checked in the same environment. ### Does this PR introduce _any_ user-facing change? No ### How was this patch tested? By running the benchmark and checking its output. Closes #28440 from MaxGekk/SPARK-31527-DateTimeBenchmark-followup. Authored-by: Max Gekk <max.gekk@gmail.com> Signed-off-by: Takeshi Yamamuro <yamamuro@apache.org>
2020-05-03 20:39:50 -04:00
OpenJDK 64-Bit Server VM 11.0.7+10-post-Ubuntu-2ubuntu218.04 on Linux 4.15.0-1063-aws
Intel(R) Xeon(R) CPU E5-2670 v2 @ 2.50GHz
[SPARK-29320][TESTS] Compare `sql/core` module in JDK8/11 (Part 1) ### What changes were proposed in this pull request? This PR regenerates the `sql/core` benchmarks in JDK8/11 to compare the result. In general, we compare the ratio instead of the time. However, in this PR, the average time is compared. This PR should be considered as a rough comparison. **A. EXPECTED CASES(JDK11 is faster in general)** - [x] BloomFilterBenchmark (JDK11 is faster except one case) - [x] BuiltInDataSourceWriteBenchmark (JDK11 is faster at CSV/ORC) - [x] CSVBenchmark (JDK11 is faster except five cases) - [x] ColumnarBatchBenchmark (JDK11 is faster at `boolean`/`string` and some cases in `int`/`array`) - [x] DatasetBenchmark (JDK11 is faster with `string`, but is slower for `long` type) - [x] ExternalAppendOnlyUnsafeRowArrayBenchmark (JDK11 is faster except two cases) - [x] ExtractBenchmark (JDK11 is faster except HOUR/MINUTE/SECOND/MILLISECONDS/MICROSECONDS) - [x] HashedRelationMetricsBenchmark (JDK11 is faster) - [x] JSONBenchmark (JDK11 is much faster except eight cases) - [x] JoinBenchmark (JDK11 is faster except five cases) - [x] OrcNestedSchemaPruningBenchmark (JDK11 is faster in nine cases) - [x] PrimitiveArrayBenchmark (JDK11 is faster) - [x] SortBenchmark (JDK11 is faster except `Arrays.sort` case) - [x] UDFBenchmark (N/A, values are too small) - [x] UnsafeArrayDataBenchmark (JDK11 is faster except one case) - [x] WideTableBenchmark (JDK11 is faster except two cases) **B. CASES WE NEED TO INVESTIGATE MORE LATER** - [x] AggregateBenchmark (JDK11 is slower in general) - [x] CompressionSchemeBenchmark (JDK11 is slower in general except `string`) - [x] DataSourceReadBenchmark (JDK11 is slower in general) - [x] DateTimeBenchmark (JDK11 is slightly slower in general except `parsing`) - [x] MakeDateTimeBenchmark (JDK11 is slower except two cases) - [x] MiscBenchmark (JDK11 is slower except ten cases) - [x] OrcV2NestedSchemaPruningBenchmark (JDK11 is slower) - [x] ParquetNestedSchemaPruningBenchmark (JDK11 is slower except six cases) - [x] RangeBenchmark (JDK11 is slower except one case) `FilterPushdownBenchmark/InExpressionBenchmark/WideSchemaBenchmark` will be compared later because it took long timer. ### Why are the changes needed? According to the result, there are some difference between JDK8/JDK11. This will be a baseline for the future improvement and comparison. Also, as a reproducible environment, the following environment is used. - Instance: `r3.xlarge` - OS: `CentOS Linux release 7.5.1804 (Core)` - JDK: - `OpenJDK Runtime Environment (build 1.8.0_222-b10)` - `OpenJDK Runtime Environment 18.9 (build 11.0.4+11-LTS)` ### Does this PR introduce any user-facing change? No. ### How was this patch tested? This is a test-only PR. We need to run benchmark. Closes #26003 from dongjoon-hyun/SPARK-29320. Authored-by: Dongjoon Hyun <dhyun@apple.com> Signed-off-by: Dongjoon Hyun <dhyun@apple.com>
2019-10-03 11:58:25 -04:00
date_trunc QUARTER: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative
------------------------------------------------------------------------------------------------------------------------
[SPARK-31630][SQL] Fix perf regression by skipping timestamps rebasing after some threshold ### What changes were proposed in this pull request? Skip timestamps rebasing after a global threshold when there is no difference between Julian and Gregorian calendars. This allows to avoid checking hash maps of switch points, and fixes perf regressions in `toJavaTimestamp()` and `fromJavaTimestamp()`. ### Why are the changes needed? The changes fix perf regressions of conversions to/from external type `java.sql.Timestamp`. Before (see the PR's results https://github.com/apache/spark/pull/28440): ``` ================================================================================================ Conversion from/to external types ================================================================================================ OpenJDK 64-Bit Server VM 1.8.0_252-8u252-b09-1~18.04-b09 on Linux 4.15.0-1063-aws Intel(R) Xeon(R) CPU E5-2670 v2 2.50GHz To/from Java's date-time: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative ------------------------------------------------------------------------------------------------------------------------ From java.sql.Timestamp 376 388 10 13.3 75.2 1.1X Collect java.sql.Timestamp 1878 1937 64 2.7 375.6 0.2X ``` After: ``` ================================================================================================ Conversion from/to external types ================================================================================================ OpenJDK 64-Bit Server VM 1.8.0_252-8u252-b09-1~18.04-b09 on Linux 4.15.0-1063-aws Intel(R) Xeon(R) CPU E5-2670 v2 2.50GHz To/from Java's date-time: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative ------------------------------------------------------------------------------------------------------------------------ From java.sql.Timestamp 249 264 24 20.1 49.8 1.7X Collect java.sql.Timestamp 1503 1523 24 3.3 300.5 0.3X ``` Perf improvements in average of: 1. From java.sql.Timestamp is ~ 34% 2. To java.sql.Timestamps is ~16% ### Does this PR introduce _any_ user-facing change? No ### How was this patch tested? By existing test suites `DateTimeUtilsSuite` and `RebaseDateTimeSuite`. Closes #28441 from MaxGekk/opt-rebase-common-threshold. Authored-by: Max Gekk <max.gekk@gmail.com> Signed-off-by: Wenchen Fan <wenchen@databricks.com>
2020-05-05 10:11:53 -04:00
date_trunc QUARTER wholestage off 3290 3292 4 3.0 329.0 1.0X
date_trunc QUARTER wholestage on 3214 3218 3 3.1 321.4 1.0X
[SPARK-29320][TESTS] Compare `sql/core` module in JDK8/11 (Part 1) ### What changes were proposed in this pull request? This PR regenerates the `sql/core` benchmarks in JDK8/11 to compare the result. In general, we compare the ratio instead of the time. However, in this PR, the average time is compared. This PR should be considered as a rough comparison. **A. EXPECTED CASES(JDK11 is faster in general)** - [x] BloomFilterBenchmark (JDK11 is faster except one case) - [x] BuiltInDataSourceWriteBenchmark (JDK11 is faster at CSV/ORC) - [x] CSVBenchmark (JDK11 is faster except five cases) - [x] ColumnarBatchBenchmark (JDK11 is faster at `boolean`/`string` and some cases in `int`/`array`) - [x] DatasetBenchmark (JDK11 is faster with `string`, but is slower for `long` type) - [x] ExternalAppendOnlyUnsafeRowArrayBenchmark (JDK11 is faster except two cases) - [x] ExtractBenchmark (JDK11 is faster except HOUR/MINUTE/SECOND/MILLISECONDS/MICROSECONDS) - [x] HashedRelationMetricsBenchmark (JDK11 is faster) - [x] JSONBenchmark (JDK11 is much faster except eight cases) - [x] JoinBenchmark (JDK11 is faster except five cases) - [x] OrcNestedSchemaPruningBenchmark (JDK11 is faster in nine cases) - [x] PrimitiveArrayBenchmark (JDK11 is faster) - [x] SortBenchmark (JDK11 is faster except `Arrays.sort` case) - [x] UDFBenchmark (N/A, values are too small) - [x] UnsafeArrayDataBenchmark (JDK11 is faster except one case) - [x] WideTableBenchmark (JDK11 is faster except two cases) **B. CASES WE NEED TO INVESTIGATE MORE LATER** - [x] AggregateBenchmark (JDK11 is slower in general) - [x] CompressionSchemeBenchmark (JDK11 is slower in general except `string`) - [x] DataSourceReadBenchmark (JDK11 is slower in general) - [x] DateTimeBenchmark (JDK11 is slightly slower in general except `parsing`) - [x] MakeDateTimeBenchmark (JDK11 is slower except two cases) - [x] MiscBenchmark (JDK11 is slower except ten cases) - [x] OrcV2NestedSchemaPruningBenchmark (JDK11 is slower) - [x] ParquetNestedSchemaPruningBenchmark (JDK11 is slower except six cases) - [x] RangeBenchmark (JDK11 is slower except one case) `FilterPushdownBenchmark/InExpressionBenchmark/WideSchemaBenchmark` will be compared later because it took long timer. ### Why are the changes needed? According to the result, there are some difference between JDK8/JDK11. This will be a baseline for the future improvement and comparison. Also, as a reproducible environment, the following environment is used. - Instance: `r3.xlarge` - OS: `CentOS Linux release 7.5.1804 (Core)` - JDK: - `OpenJDK Runtime Environment (build 1.8.0_222-b10)` - `OpenJDK Runtime Environment 18.9 (build 11.0.4+11-LTS)` ### Does this PR introduce any user-facing change? No. ### How was this patch tested? This is a test-only PR. We need to run benchmark. Closes #26003 from dongjoon-hyun/SPARK-29320. Authored-by: Dongjoon Hyun <dhyun@apple.com> Signed-off-by: Dongjoon Hyun <dhyun@apple.com>
2019-10-03 11:58:25 -04:00
[SPARK-31527][SQL][TESTS][FOLLOWUP] Fix the number of rows in `DateTimeBenchmark` ### What changes were proposed in this pull request? - Changed to the number of rows in benchmark cases from 3 to the actual number `N`. - Regenerated benchmark results in the environment: | Item | Description | | ---- | ----| | Region | us-west-2 (Oregon) | | Instance | r3.xlarge | | AMI | ubuntu/images/hvm-ssd/ubuntu-bionic-18.04-amd64-server-20190722.1 (ami-06f2f779464715dc5) | | Java | OpenJDK 64-Bit Server VM 1.8.0_242 and OpenJDK 64-Bit Server VM 11.0.6+10 | ### Why are the changes needed? The changes are needed to have: - Correct benchmark results - Base line for other perf improvements that can be checked in the same environment. ### Does this PR introduce _any_ user-facing change? No ### How was this patch tested? By running the benchmark and checking its output. Closes #28440 from MaxGekk/SPARK-31527-DateTimeBenchmark-followup. Authored-by: Max Gekk <max.gekk@gmail.com> Signed-off-by: Takeshi Yamamuro <yamamuro@apache.org>
2020-05-03 20:39:50 -04:00
OpenJDK 64-Bit Server VM 11.0.7+10-post-Ubuntu-2ubuntu218.04 on Linux 4.15.0-1063-aws
Intel(R) Xeon(R) CPU E5-2670 v2 @ 2.50GHz
[SPARK-29320][TESTS] Compare `sql/core` module in JDK8/11 (Part 1) ### What changes were proposed in this pull request? This PR regenerates the `sql/core` benchmarks in JDK8/11 to compare the result. In general, we compare the ratio instead of the time. However, in this PR, the average time is compared. This PR should be considered as a rough comparison. **A. EXPECTED CASES(JDK11 is faster in general)** - [x] BloomFilterBenchmark (JDK11 is faster except one case) - [x] BuiltInDataSourceWriteBenchmark (JDK11 is faster at CSV/ORC) - [x] CSVBenchmark (JDK11 is faster except five cases) - [x] ColumnarBatchBenchmark (JDK11 is faster at `boolean`/`string` and some cases in `int`/`array`) - [x] DatasetBenchmark (JDK11 is faster with `string`, but is slower for `long` type) - [x] ExternalAppendOnlyUnsafeRowArrayBenchmark (JDK11 is faster except two cases) - [x] ExtractBenchmark (JDK11 is faster except HOUR/MINUTE/SECOND/MILLISECONDS/MICROSECONDS) - [x] HashedRelationMetricsBenchmark (JDK11 is faster) - [x] JSONBenchmark (JDK11 is much faster except eight cases) - [x] JoinBenchmark (JDK11 is faster except five cases) - [x] OrcNestedSchemaPruningBenchmark (JDK11 is faster in nine cases) - [x] PrimitiveArrayBenchmark (JDK11 is faster) - [x] SortBenchmark (JDK11 is faster except `Arrays.sort` case) - [x] UDFBenchmark (N/A, values are too small) - [x] UnsafeArrayDataBenchmark (JDK11 is faster except one case) - [x] WideTableBenchmark (JDK11 is faster except two cases) **B. CASES WE NEED TO INVESTIGATE MORE LATER** - [x] AggregateBenchmark (JDK11 is slower in general) - [x] CompressionSchemeBenchmark (JDK11 is slower in general except `string`) - [x] DataSourceReadBenchmark (JDK11 is slower in general) - [x] DateTimeBenchmark (JDK11 is slightly slower in general except `parsing`) - [x] MakeDateTimeBenchmark (JDK11 is slower except two cases) - [x] MiscBenchmark (JDK11 is slower except ten cases) - [x] OrcV2NestedSchemaPruningBenchmark (JDK11 is slower) - [x] ParquetNestedSchemaPruningBenchmark (JDK11 is slower except six cases) - [x] RangeBenchmark (JDK11 is slower except one case) `FilterPushdownBenchmark/InExpressionBenchmark/WideSchemaBenchmark` will be compared later because it took long timer. ### Why are the changes needed? According to the result, there are some difference between JDK8/JDK11. This will be a baseline for the future improvement and comparison. Also, as a reproducible environment, the following environment is used. - Instance: `r3.xlarge` - OS: `CentOS Linux release 7.5.1804 (Core)` - JDK: - `OpenJDK Runtime Environment (build 1.8.0_222-b10)` - `OpenJDK Runtime Environment 18.9 (build 11.0.4+11-LTS)` ### Does this PR introduce any user-facing change? No. ### How was this patch tested? This is a test-only PR. We need to run benchmark. Closes #26003 from dongjoon-hyun/SPARK-29320. Authored-by: Dongjoon Hyun <dhyun@apple.com> Signed-off-by: Dongjoon Hyun <dhyun@apple.com>
2019-10-03 11:58:25 -04:00
trunc year: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative
------------------------------------------------------------------------------------------------------------------------
[SPARK-31630][SQL] Fix perf regression by skipping timestamps rebasing after some threshold ### What changes were proposed in this pull request? Skip timestamps rebasing after a global threshold when there is no difference between Julian and Gregorian calendars. This allows to avoid checking hash maps of switch points, and fixes perf regressions in `toJavaTimestamp()` and `fromJavaTimestamp()`. ### Why are the changes needed? The changes fix perf regressions of conversions to/from external type `java.sql.Timestamp`. Before (see the PR's results https://github.com/apache/spark/pull/28440): ``` ================================================================================================ Conversion from/to external types ================================================================================================ OpenJDK 64-Bit Server VM 1.8.0_252-8u252-b09-1~18.04-b09 on Linux 4.15.0-1063-aws Intel(R) Xeon(R) CPU E5-2670 v2 2.50GHz To/from Java's date-time: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative ------------------------------------------------------------------------------------------------------------------------ From java.sql.Timestamp 376 388 10 13.3 75.2 1.1X Collect java.sql.Timestamp 1878 1937 64 2.7 375.6 0.2X ``` After: ``` ================================================================================================ Conversion from/to external types ================================================================================================ OpenJDK 64-Bit Server VM 1.8.0_252-8u252-b09-1~18.04-b09 on Linux 4.15.0-1063-aws Intel(R) Xeon(R) CPU E5-2670 v2 2.50GHz To/from Java's date-time: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative ------------------------------------------------------------------------------------------------------------------------ From java.sql.Timestamp 249 264 24 20.1 49.8 1.7X Collect java.sql.Timestamp 1503 1523 24 3.3 300.5 0.3X ``` Perf improvements in average of: 1. From java.sql.Timestamp is ~ 34% 2. To java.sql.Timestamps is ~16% ### Does this PR introduce _any_ user-facing change? No ### How was this patch tested? By existing test suites `DateTimeUtilsSuite` and `RebaseDateTimeSuite`. Closes #28441 from MaxGekk/opt-rebase-common-threshold. Authored-by: Max Gekk <max.gekk@gmail.com> Signed-off-by: Wenchen Fan <wenchen@databricks.com>
2020-05-05 10:11:53 -04:00
trunc year wholestage off 308 310 3 32.5 30.8 1.0X
trunc year wholestage on 289 293 6 34.7 28.9 1.1X
[SPARK-29320][TESTS] Compare `sql/core` module in JDK8/11 (Part 1) ### What changes were proposed in this pull request? This PR regenerates the `sql/core` benchmarks in JDK8/11 to compare the result. In general, we compare the ratio instead of the time. However, in this PR, the average time is compared. This PR should be considered as a rough comparison. **A. EXPECTED CASES(JDK11 is faster in general)** - [x] BloomFilterBenchmark (JDK11 is faster except one case) - [x] BuiltInDataSourceWriteBenchmark (JDK11 is faster at CSV/ORC) - [x] CSVBenchmark (JDK11 is faster except five cases) - [x] ColumnarBatchBenchmark (JDK11 is faster at `boolean`/`string` and some cases in `int`/`array`) - [x] DatasetBenchmark (JDK11 is faster with `string`, but is slower for `long` type) - [x] ExternalAppendOnlyUnsafeRowArrayBenchmark (JDK11 is faster except two cases) - [x] ExtractBenchmark (JDK11 is faster except HOUR/MINUTE/SECOND/MILLISECONDS/MICROSECONDS) - [x] HashedRelationMetricsBenchmark (JDK11 is faster) - [x] JSONBenchmark (JDK11 is much faster except eight cases) - [x] JoinBenchmark (JDK11 is faster except five cases) - [x] OrcNestedSchemaPruningBenchmark (JDK11 is faster in nine cases) - [x] PrimitiveArrayBenchmark (JDK11 is faster) - [x] SortBenchmark (JDK11 is faster except `Arrays.sort` case) - [x] UDFBenchmark (N/A, values are too small) - [x] UnsafeArrayDataBenchmark (JDK11 is faster except one case) - [x] WideTableBenchmark (JDK11 is faster except two cases) **B. CASES WE NEED TO INVESTIGATE MORE LATER** - [x] AggregateBenchmark (JDK11 is slower in general) - [x] CompressionSchemeBenchmark (JDK11 is slower in general except `string`) - [x] DataSourceReadBenchmark (JDK11 is slower in general) - [x] DateTimeBenchmark (JDK11 is slightly slower in general except `parsing`) - [x] MakeDateTimeBenchmark (JDK11 is slower except two cases) - [x] MiscBenchmark (JDK11 is slower except ten cases) - [x] OrcV2NestedSchemaPruningBenchmark (JDK11 is slower) - [x] ParquetNestedSchemaPruningBenchmark (JDK11 is slower except six cases) - [x] RangeBenchmark (JDK11 is slower except one case) `FilterPushdownBenchmark/InExpressionBenchmark/WideSchemaBenchmark` will be compared later because it took long timer. ### Why are the changes needed? According to the result, there are some difference between JDK8/JDK11. This will be a baseline for the future improvement and comparison. Also, as a reproducible environment, the following environment is used. - Instance: `r3.xlarge` - OS: `CentOS Linux release 7.5.1804 (Core)` - JDK: - `OpenJDK Runtime Environment (build 1.8.0_222-b10)` - `OpenJDK Runtime Environment 18.9 (build 11.0.4+11-LTS)` ### Does this PR introduce any user-facing change? No. ### How was this patch tested? This is a test-only PR. We need to run benchmark. Closes #26003 from dongjoon-hyun/SPARK-29320. Authored-by: Dongjoon Hyun <dhyun@apple.com> Signed-off-by: Dongjoon Hyun <dhyun@apple.com>
2019-10-03 11:58:25 -04:00
[SPARK-31527][SQL][TESTS][FOLLOWUP] Fix the number of rows in `DateTimeBenchmark` ### What changes were proposed in this pull request? - Changed to the number of rows in benchmark cases from 3 to the actual number `N`. - Regenerated benchmark results in the environment: | Item | Description | | ---- | ----| | Region | us-west-2 (Oregon) | | Instance | r3.xlarge | | AMI | ubuntu/images/hvm-ssd/ubuntu-bionic-18.04-amd64-server-20190722.1 (ami-06f2f779464715dc5) | | Java | OpenJDK 64-Bit Server VM 1.8.0_242 and OpenJDK 64-Bit Server VM 11.0.6+10 | ### Why are the changes needed? The changes are needed to have: - Correct benchmark results - Base line for other perf improvements that can be checked in the same environment. ### Does this PR introduce _any_ user-facing change? No ### How was this patch tested? By running the benchmark and checking its output. Closes #28440 from MaxGekk/SPARK-31527-DateTimeBenchmark-followup. Authored-by: Max Gekk <max.gekk@gmail.com> Signed-off-by: Takeshi Yamamuro <yamamuro@apache.org>
2020-05-03 20:39:50 -04:00
OpenJDK 64-Bit Server VM 11.0.7+10-post-Ubuntu-2ubuntu218.04 on Linux 4.15.0-1063-aws
Intel(R) Xeon(R) CPU E5-2670 v2 @ 2.50GHz
[SPARK-29320][TESTS] Compare `sql/core` module in JDK8/11 (Part 1) ### What changes were proposed in this pull request? This PR regenerates the `sql/core` benchmarks in JDK8/11 to compare the result. In general, we compare the ratio instead of the time. However, in this PR, the average time is compared. This PR should be considered as a rough comparison. **A. EXPECTED CASES(JDK11 is faster in general)** - [x] BloomFilterBenchmark (JDK11 is faster except one case) - [x] BuiltInDataSourceWriteBenchmark (JDK11 is faster at CSV/ORC) - [x] CSVBenchmark (JDK11 is faster except five cases) - [x] ColumnarBatchBenchmark (JDK11 is faster at `boolean`/`string` and some cases in `int`/`array`) - [x] DatasetBenchmark (JDK11 is faster with `string`, but is slower for `long` type) - [x] ExternalAppendOnlyUnsafeRowArrayBenchmark (JDK11 is faster except two cases) - [x] ExtractBenchmark (JDK11 is faster except HOUR/MINUTE/SECOND/MILLISECONDS/MICROSECONDS) - [x] HashedRelationMetricsBenchmark (JDK11 is faster) - [x] JSONBenchmark (JDK11 is much faster except eight cases) - [x] JoinBenchmark (JDK11 is faster except five cases) - [x] OrcNestedSchemaPruningBenchmark (JDK11 is faster in nine cases) - [x] PrimitiveArrayBenchmark (JDK11 is faster) - [x] SortBenchmark (JDK11 is faster except `Arrays.sort` case) - [x] UDFBenchmark (N/A, values are too small) - [x] UnsafeArrayDataBenchmark (JDK11 is faster except one case) - [x] WideTableBenchmark (JDK11 is faster except two cases) **B. CASES WE NEED TO INVESTIGATE MORE LATER** - [x] AggregateBenchmark (JDK11 is slower in general) - [x] CompressionSchemeBenchmark (JDK11 is slower in general except `string`) - [x] DataSourceReadBenchmark (JDK11 is slower in general) - [x] DateTimeBenchmark (JDK11 is slightly slower in general except `parsing`) - [x] MakeDateTimeBenchmark (JDK11 is slower except two cases) - [x] MiscBenchmark (JDK11 is slower except ten cases) - [x] OrcV2NestedSchemaPruningBenchmark (JDK11 is slower) - [x] ParquetNestedSchemaPruningBenchmark (JDK11 is slower except six cases) - [x] RangeBenchmark (JDK11 is slower except one case) `FilterPushdownBenchmark/InExpressionBenchmark/WideSchemaBenchmark` will be compared later because it took long timer. ### Why are the changes needed? According to the result, there are some difference between JDK8/JDK11. This will be a baseline for the future improvement and comparison. Also, as a reproducible environment, the following environment is used. - Instance: `r3.xlarge` - OS: `CentOS Linux release 7.5.1804 (Core)` - JDK: - `OpenJDK Runtime Environment (build 1.8.0_222-b10)` - `OpenJDK Runtime Environment 18.9 (build 11.0.4+11-LTS)` ### Does this PR introduce any user-facing change? No. ### How was this patch tested? This is a test-only PR. We need to run benchmark. Closes #26003 from dongjoon-hyun/SPARK-29320. Authored-by: Dongjoon Hyun <dhyun@apple.com> Signed-off-by: Dongjoon Hyun <dhyun@apple.com>
2019-10-03 11:58:25 -04:00
trunc yyyy: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative
------------------------------------------------------------------------------------------------------------------------
[SPARK-31630][SQL] Fix perf regression by skipping timestamps rebasing after some threshold ### What changes were proposed in this pull request? Skip timestamps rebasing after a global threshold when there is no difference between Julian and Gregorian calendars. This allows to avoid checking hash maps of switch points, and fixes perf regressions in `toJavaTimestamp()` and `fromJavaTimestamp()`. ### Why are the changes needed? The changes fix perf regressions of conversions to/from external type `java.sql.Timestamp`. Before (see the PR's results https://github.com/apache/spark/pull/28440): ``` ================================================================================================ Conversion from/to external types ================================================================================================ OpenJDK 64-Bit Server VM 1.8.0_252-8u252-b09-1~18.04-b09 on Linux 4.15.0-1063-aws Intel(R) Xeon(R) CPU E5-2670 v2 2.50GHz To/from Java's date-time: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative ------------------------------------------------------------------------------------------------------------------------ From java.sql.Timestamp 376 388 10 13.3 75.2 1.1X Collect java.sql.Timestamp 1878 1937 64 2.7 375.6 0.2X ``` After: ``` ================================================================================================ Conversion from/to external types ================================================================================================ OpenJDK 64-Bit Server VM 1.8.0_252-8u252-b09-1~18.04-b09 on Linux 4.15.0-1063-aws Intel(R) Xeon(R) CPU E5-2670 v2 2.50GHz To/from Java's date-time: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative ------------------------------------------------------------------------------------------------------------------------ From java.sql.Timestamp 249 264 24 20.1 49.8 1.7X Collect java.sql.Timestamp 1503 1523 24 3.3 300.5 0.3X ``` Perf improvements in average of: 1. From java.sql.Timestamp is ~ 34% 2. To java.sql.Timestamps is ~16% ### Does this PR introduce _any_ user-facing change? No ### How was this patch tested? By existing test suites `DateTimeUtilsSuite` and `RebaseDateTimeSuite`. Closes #28441 from MaxGekk/opt-rebase-common-threshold. Authored-by: Max Gekk <max.gekk@gmail.com> Signed-off-by: Wenchen Fan <wenchen@databricks.com>
2020-05-05 10:11:53 -04:00
trunc yyyy wholestage off 309 311 3 32.4 30.9 1.0X
trunc yyyy wholestage on 289 294 7 34.6 28.9 1.1X
[SPARK-29320][TESTS] Compare `sql/core` module in JDK8/11 (Part 1) ### What changes were proposed in this pull request? This PR regenerates the `sql/core` benchmarks in JDK8/11 to compare the result. In general, we compare the ratio instead of the time. However, in this PR, the average time is compared. This PR should be considered as a rough comparison. **A. EXPECTED CASES(JDK11 is faster in general)** - [x] BloomFilterBenchmark (JDK11 is faster except one case) - [x] BuiltInDataSourceWriteBenchmark (JDK11 is faster at CSV/ORC) - [x] CSVBenchmark (JDK11 is faster except five cases) - [x] ColumnarBatchBenchmark (JDK11 is faster at `boolean`/`string` and some cases in `int`/`array`) - [x] DatasetBenchmark (JDK11 is faster with `string`, but is slower for `long` type) - [x] ExternalAppendOnlyUnsafeRowArrayBenchmark (JDK11 is faster except two cases) - [x] ExtractBenchmark (JDK11 is faster except HOUR/MINUTE/SECOND/MILLISECONDS/MICROSECONDS) - [x] HashedRelationMetricsBenchmark (JDK11 is faster) - [x] JSONBenchmark (JDK11 is much faster except eight cases) - [x] JoinBenchmark (JDK11 is faster except five cases) - [x] OrcNestedSchemaPruningBenchmark (JDK11 is faster in nine cases) - [x] PrimitiveArrayBenchmark (JDK11 is faster) - [x] SortBenchmark (JDK11 is faster except `Arrays.sort` case) - [x] UDFBenchmark (N/A, values are too small) - [x] UnsafeArrayDataBenchmark (JDK11 is faster except one case) - [x] WideTableBenchmark (JDK11 is faster except two cases) **B. CASES WE NEED TO INVESTIGATE MORE LATER** - [x] AggregateBenchmark (JDK11 is slower in general) - [x] CompressionSchemeBenchmark (JDK11 is slower in general except `string`) - [x] DataSourceReadBenchmark (JDK11 is slower in general) - [x] DateTimeBenchmark (JDK11 is slightly slower in general except `parsing`) - [x] MakeDateTimeBenchmark (JDK11 is slower except two cases) - [x] MiscBenchmark (JDK11 is slower except ten cases) - [x] OrcV2NestedSchemaPruningBenchmark (JDK11 is slower) - [x] ParquetNestedSchemaPruningBenchmark (JDK11 is slower except six cases) - [x] RangeBenchmark (JDK11 is slower except one case) `FilterPushdownBenchmark/InExpressionBenchmark/WideSchemaBenchmark` will be compared later because it took long timer. ### Why are the changes needed? According to the result, there are some difference between JDK8/JDK11. This will be a baseline for the future improvement and comparison. Also, as a reproducible environment, the following environment is used. - Instance: `r3.xlarge` - OS: `CentOS Linux release 7.5.1804 (Core)` - JDK: - `OpenJDK Runtime Environment (build 1.8.0_222-b10)` - `OpenJDK Runtime Environment 18.9 (build 11.0.4+11-LTS)` ### Does this PR introduce any user-facing change? No. ### How was this patch tested? This is a test-only PR. We need to run benchmark. Closes #26003 from dongjoon-hyun/SPARK-29320. Authored-by: Dongjoon Hyun <dhyun@apple.com> Signed-off-by: Dongjoon Hyun <dhyun@apple.com>
2019-10-03 11:58:25 -04:00
[SPARK-31527][SQL][TESTS][FOLLOWUP] Fix the number of rows in `DateTimeBenchmark` ### What changes were proposed in this pull request? - Changed to the number of rows in benchmark cases from 3 to the actual number `N`. - Regenerated benchmark results in the environment: | Item | Description | | ---- | ----| | Region | us-west-2 (Oregon) | | Instance | r3.xlarge | | AMI | ubuntu/images/hvm-ssd/ubuntu-bionic-18.04-amd64-server-20190722.1 (ami-06f2f779464715dc5) | | Java | OpenJDK 64-Bit Server VM 1.8.0_242 and OpenJDK 64-Bit Server VM 11.0.6+10 | ### Why are the changes needed? The changes are needed to have: - Correct benchmark results - Base line for other perf improvements that can be checked in the same environment. ### Does this PR introduce _any_ user-facing change? No ### How was this patch tested? By running the benchmark and checking its output. Closes #28440 from MaxGekk/SPARK-31527-DateTimeBenchmark-followup. Authored-by: Max Gekk <max.gekk@gmail.com> Signed-off-by: Takeshi Yamamuro <yamamuro@apache.org>
2020-05-03 20:39:50 -04:00
OpenJDK 64-Bit Server VM 11.0.7+10-post-Ubuntu-2ubuntu218.04 on Linux 4.15.0-1063-aws
Intel(R) Xeon(R) CPU E5-2670 v2 @ 2.50GHz
[SPARK-29320][TESTS] Compare `sql/core` module in JDK8/11 (Part 1) ### What changes were proposed in this pull request? This PR regenerates the `sql/core` benchmarks in JDK8/11 to compare the result. In general, we compare the ratio instead of the time. However, in this PR, the average time is compared. This PR should be considered as a rough comparison. **A. EXPECTED CASES(JDK11 is faster in general)** - [x] BloomFilterBenchmark (JDK11 is faster except one case) - [x] BuiltInDataSourceWriteBenchmark (JDK11 is faster at CSV/ORC) - [x] CSVBenchmark (JDK11 is faster except five cases) - [x] ColumnarBatchBenchmark (JDK11 is faster at `boolean`/`string` and some cases in `int`/`array`) - [x] DatasetBenchmark (JDK11 is faster with `string`, but is slower for `long` type) - [x] ExternalAppendOnlyUnsafeRowArrayBenchmark (JDK11 is faster except two cases) - [x] ExtractBenchmark (JDK11 is faster except HOUR/MINUTE/SECOND/MILLISECONDS/MICROSECONDS) - [x] HashedRelationMetricsBenchmark (JDK11 is faster) - [x] JSONBenchmark (JDK11 is much faster except eight cases) - [x] JoinBenchmark (JDK11 is faster except five cases) - [x] OrcNestedSchemaPruningBenchmark (JDK11 is faster in nine cases) - [x] PrimitiveArrayBenchmark (JDK11 is faster) - [x] SortBenchmark (JDK11 is faster except `Arrays.sort` case) - [x] UDFBenchmark (N/A, values are too small) - [x] UnsafeArrayDataBenchmark (JDK11 is faster except one case) - [x] WideTableBenchmark (JDK11 is faster except two cases) **B. CASES WE NEED TO INVESTIGATE MORE LATER** - [x] AggregateBenchmark (JDK11 is slower in general) - [x] CompressionSchemeBenchmark (JDK11 is slower in general except `string`) - [x] DataSourceReadBenchmark (JDK11 is slower in general) - [x] DateTimeBenchmark (JDK11 is slightly slower in general except `parsing`) - [x] MakeDateTimeBenchmark (JDK11 is slower except two cases) - [x] MiscBenchmark (JDK11 is slower except ten cases) - [x] OrcV2NestedSchemaPruningBenchmark (JDK11 is slower) - [x] ParquetNestedSchemaPruningBenchmark (JDK11 is slower except six cases) - [x] RangeBenchmark (JDK11 is slower except one case) `FilterPushdownBenchmark/InExpressionBenchmark/WideSchemaBenchmark` will be compared later because it took long timer. ### Why are the changes needed? According to the result, there are some difference between JDK8/JDK11. This will be a baseline for the future improvement and comparison. Also, as a reproducible environment, the following environment is used. - Instance: `r3.xlarge` - OS: `CentOS Linux release 7.5.1804 (Core)` - JDK: - `OpenJDK Runtime Environment (build 1.8.0_222-b10)` - `OpenJDK Runtime Environment 18.9 (build 11.0.4+11-LTS)` ### Does this PR introduce any user-facing change? No. ### How was this patch tested? This is a test-only PR. We need to run benchmark. Closes #26003 from dongjoon-hyun/SPARK-29320. Authored-by: Dongjoon Hyun <dhyun@apple.com> Signed-off-by: Dongjoon Hyun <dhyun@apple.com>
2019-10-03 11:58:25 -04:00
trunc yy: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative
------------------------------------------------------------------------------------------------------------------------
[SPARK-31630][SQL] Fix perf regression by skipping timestamps rebasing after some threshold ### What changes were proposed in this pull request? Skip timestamps rebasing after a global threshold when there is no difference between Julian and Gregorian calendars. This allows to avoid checking hash maps of switch points, and fixes perf regressions in `toJavaTimestamp()` and `fromJavaTimestamp()`. ### Why are the changes needed? The changes fix perf regressions of conversions to/from external type `java.sql.Timestamp`. Before (see the PR's results https://github.com/apache/spark/pull/28440): ``` ================================================================================================ Conversion from/to external types ================================================================================================ OpenJDK 64-Bit Server VM 1.8.0_252-8u252-b09-1~18.04-b09 on Linux 4.15.0-1063-aws Intel(R) Xeon(R) CPU E5-2670 v2 2.50GHz To/from Java's date-time: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative ------------------------------------------------------------------------------------------------------------------------ From java.sql.Timestamp 376 388 10 13.3 75.2 1.1X Collect java.sql.Timestamp 1878 1937 64 2.7 375.6 0.2X ``` After: ``` ================================================================================================ Conversion from/to external types ================================================================================================ OpenJDK 64-Bit Server VM 1.8.0_252-8u252-b09-1~18.04-b09 on Linux 4.15.0-1063-aws Intel(R) Xeon(R) CPU E5-2670 v2 2.50GHz To/from Java's date-time: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative ------------------------------------------------------------------------------------------------------------------------ From java.sql.Timestamp 249 264 24 20.1 49.8 1.7X Collect java.sql.Timestamp 1503 1523 24 3.3 300.5 0.3X ``` Perf improvements in average of: 1. From java.sql.Timestamp is ~ 34% 2. To java.sql.Timestamps is ~16% ### Does this PR introduce _any_ user-facing change? No ### How was this patch tested? By existing test suites `DateTimeUtilsSuite` and `RebaseDateTimeSuite`. Closes #28441 from MaxGekk/opt-rebase-common-threshold. Authored-by: Max Gekk <max.gekk@gmail.com> Signed-off-by: Wenchen Fan <wenchen@databricks.com>
2020-05-05 10:11:53 -04:00
trunc yy wholestage off 311 311 0 32.2 31.1 1.0X
trunc yy wholestage on 288 294 7 34.7 28.8 1.1X
[SPARK-29320][TESTS] Compare `sql/core` module in JDK8/11 (Part 1) ### What changes were proposed in this pull request? This PR regenerates the `sql/core` benchmarks in JDK8/11 to compare the result. In general, we compare the ratio instead of the time. However, in this PR, the average time is compared. This PR should be considered as a rough comparison. **A. EXPECTED CASES(JDK11 is faster in general)** - [x] BloomFilterBenchmark (JDK11 is faster except one case) - [x] BuiltInDataSourceWriteBenchmark (JDK11 is faster at CSV/ORC) - [x] CSVBenchmark (JDK11 is faster except five cases) - [x] ColumnarBatchBenchmark (JDK11 is faster at `boolean`/`string` and some cases in `int`/`array`) - [x] DatasetBenchmark (JDK11 is faster with `string`, but is slower for `long` type) - [x] ExternalAppendOnlyUnsafeRowArrayBenchmark (JDK11 is faster except two cases) - [x] ExtractBenchmark (JDK11 is faster except HOUR/MINUTE/SECOND/MILLISECONDS/MICROSECONDS) - [x] HashedRelationMetricsBenchmark (JDK11 is faster) - [x] JSONBenchmark (JDK11 is much faster except eight cases) - [x] JoinBenchmark (JDK11 is faster except five cases) - [x] OrcNestedSchemaPruningBenchmark (JDK11 is faster in nine cases) - [x] PrimitiveArrayBenchmark (JDK11 is faster) - [x] SortBenchmark (JDK11 is faster except `Arrays.sort` case) - [x] UDFBenchmark (N/A, values are too small) - [x] UnsafeArrayDataBenchmark (JDK11 is faster except one case) - [x] WideTableBenchmark (JDK11 is faster except two cases) **B. CASES WE NEED TO INVESTIGATE MORE LATER** - [x] AggregateBenchmark (JDK11 is slower in general) - [x] CompressionSchemeBenchmark (JDK11 is slower in general except `string`) - [x] DataSourceReadBenchmark (JDK11 is slower in general) - [x] DateTimeBenchmark (JDK11 is slightly slower in general except `parsing`) - [x] MakeDateTimeBenchmark (JDK11 is slower except two cases) - [x] MiscBenchmark (JDK11 is slower except ten cases) - [x] OrcV2NestedSchemaPruningBenchmark (JDK11 is slower) - [x] ParquetNestedSchemaPruningBenchmark (JDK11 is slower except six cases) - [x] RangeBenchmark (JDK11 is slower except one case) `FilterPushdownBenchmark/InExpressionBenchmark/WideSchemaBenchmark` will be compared later because it took long timer. ### Why are the changes needed? According to the result, there are some difference between JDK8/JDK11. This will be a baseline for the future improvement and comparison. Also, as a reproducible environment, the following environment is used. - Instance: `r3.xlarge` - OS: `CentOS Linux release 7.5.1804 (Core)` - JDK: - `OpenJDK Runtime Environment (build 1.8.0_222-b10)` - `OpenJDK Runtime Environment 18.9 (build 11.0.4+11-LTS)` ### Does this PR introduce any user-facing change? No. ### How was this patch tested? This is a test-only PR. We need to run benchmark. Closes #26003 from dongjoon-hyun/SPARK-29320. Authored-by: Dongjoon Hyun <dhyun@apple.com> Signed-off-by: Dongjoon Hyun <dhyun@apple.com>
2019-10-03 11:58:25 -04:00
[SPARK-31527][SQL][TESTS][FOLLOWUP] Fix the number of rows in `DateTimeBenchmark` ### What changes were proposed in this pull request? - Changed to the number of rows in benchmark cases from 3 to the actual number `N`. - Regenerated benchmark results in the environment: | Item | Description | | ---- | ----| | Region | us-west-2 (Oregon) | | Instance | r3.xlarge | | AMI | ubuntu/images/hvm-ssd/ubuntu-bionic-18.04-amd64-server-20190722.1 (ami-06f2f779464715dc5) | | Java | OpenJDK 64-Bit Server VM 1.8.0_242 and OpenJDK 64-Bit Server VM 11.0.6+10 | ### Why are the changes needed? The changes are needed to have: - Correct benchmark results - Base line for other perf improvements that can be checked in the same environment. ### Does this PR introduce _any_ user-facing change? No ### How was this patch tested? By running the benchmark and checking its output. Closes #28440 from MaxGekk/SPARK-31527-DateTimeBenchmark-followup. Authored-by: Max Gekk <max.gekk@gmail.com> Signed-off-by: Takeshi Yamamuro <yamamuro@apache.org>
2020-05-03 20:39:50 -04:00
OpenJDK 64-Bit Server VM 11.0.7+10-post-Ubuntu-2ubuntu218.04 on Linux 4.15.0-1063-aws
Intel(R) Xeon(R) CPU E5-2670 v2 @ 2.50GHz
[SPARK-29320][TESTS] Compare `sql/core` module in JDK8/11 (Part 1) ### What changes were proposed in this pull request? This PR regenerates the `sql/core` benchmarks in JDK8/11 to compare the result. In general, we compare the ratio instead of the time. However, in this PR, the average time is compared. This PR should be considered as a rough comparison. **A. EXPECTED CASES(JDK11 is faster in general)** - [x] BloomFilterBenchmark (JDK11 is faster except one case) - [x] BuiltInDataSourceWriteBenchmark (JDK11 is faster at CSV/ORC) - [x] CSVBenchmark (JDK11 is faster except five cases) - [x] ColumnarBatchBenchmark (JDK11 is faster at `boolean`/`string` and some cases in `int`/`array`) - [x] DatasetBenchmark (JDK11 is faster with `string`, but is slower for `long` type) - [x] ExternalAppendOnlyUnsafeRowArrayBenchmark (JDK11 is faster except two cases) - [x] ExtractBenchmark (JDK11 is faster except HOUR/MINUTE/SECOND/MILLISECONDS/MICROSECONDS) - [x] HashedRelationMetricsBenchmark (JDK11 is faster) - [x] JSONBenchmark (JDK11 is much faster except eight cases) - [x] JoinBenchmark (JDK11 is faster except five cases) - [x] OrcNestedSchemaPruningBenchmark (JDK11 is faster in nine cases) - [x] PrimitiveArrayBenchmark (JDK11 is faster) - [x] SortBenchmark (JDK11 is faster except `Arrays.sort` case) - [x] UDFBenchmark (N/A, values are too small) - [x] UnsafeArrayDataBenchmark (JDK11 is faster except one case) - [x] WideTableBenchmark (JDK11 is faster except two cases) **B. CASES WE NEED TO INVESTIGATE MORE LATER** - [x] AggregateBenchmark (JDK11 is slower in general) - [x] CompressionSchemeBenchmark (JDK11 is slower in general except `string`) - [x] DataSourceReadBenchmark (JDK11 is slower in general) - [x] DateTimeBenchmark (JDK11 is slightly slower in general except `parsing`) - [x] MakeDateTimeBenchmark (JDK11 is slower except two cases) - [x] MiscBenchmark (JDK11 is slower except ten cases) - [x] OrcV2NestedSchemaPruningBenchmark (JDK11 is slower) - [x] ParquetNestedSchemaPruningBenchmark (JDK11 is slower except six cases) - [x] RangeBenchmark (JDK11 is slower except one case) `FilterPushdownBenchmark/InExpressionBenchmark/WideSchemaBenchmark` will be compared later because it took long timer. ### Why are the changes needed? According to the result, there are some difference between JDK8/JDK11. This will be a baseline for the future improvement and comparison. Also, as a reproducible environment, the following environment is used. - Instance: `r3.xlarge` - OS: `CentOS Linux release 7.5.1804 (Core)` - JDK: - `OpenJDK Runtime Environment (build 1.8.0_222-b10)` - `OpenJDK Runtime Environment 18.9 (build 11.0.4+11-LTS)` ### Does this PR introduce any user-facing change? No. ### How was this patch tested? This is a test-only PR. We need to run benchmark. Closes #26003 from dongjoon-hyun/SPARK-29320. Authored-by: Dongjoon Hyun <dhyun@apple.com> Signed-off-by: Dongjoon Hyun <dhyun@apple.com>
2019-10-03 11:58:25 -04:00
trunc mon: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative
------------------------------------------------------------------------------------------------------------------------
[SPARK-31630][SQL] Fix perf regression by skipping timestamps rebasing after some threshold ### What changes were proposed in this pull request? Skip timestamps rebasing after a global threshold when there is no difference between Julian and Gregorian calendars. This allows to avoid checking hash maps of switch points, and fixes perf regressions in `toJavaTimestamp()` and `fromJavaTimestamp()`. ### Why are the changes needed? The changes fix perf regressions of conversions to/from external type `java.sql.Timestamp`. Before (see the PR's results https://github.com/apache/spark/pull/28440): ``` ================================================================================================ Conversion from/to external types ================================================================================================ OpenJDK 64-Bit Server VM 1.8.0_252-8u252-b09-1~18.04-b09 on Linux 4.15.0-1063-aws Intel(R) Xeon(R) CPU E5-2670 v2 2.50GHz To/from Java's date-time: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative ------------------------------------------------------------------------------------------------------------------------ From java.sql.Timestamp 376 388 10 13.3 75.2 1.1X Collect java.sql.Timestamp 1878 1937 64 2.7 375.6 0.2X ``` After: ``` ================================================================================================ Conversion from/to external types ================================================================================================ OpenJDK 64-Bit Server VM 1.8.0_252-8u252-b09-1~18.04-b09 on Linux 4.15.0-1063-aws Intel(R) Xeon(R) CPU E5-2670 v2 2.50GHz To/from Java's date-time: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative ------------------------------------------------------------------------------------------------------------------------ From java.sql.Timestamp 249 264 24 20.1 49.8 1.7X Collect java.sql.Timestamp 1503 1523 24 3.3 300.5 0.3X ``` Perf improvements in average of: 1. From java.sql.Timestamp is ~ 34% 2. To java.sql.Timestamps is ~16% ### Does this PR introduce _any_ user-facing change? No ### How was this patch tested? By existing test suites `DateTimeUtilsSuite` and `RebaseDateTimeSuite`. Closes #28441 from MaxGekk/opt-rebase-common-threshold. Authored-by: Max Gekk <max.gekk@gmail.com> Signed-off-by: Wenchen Fan <wenchen@databricks.com>
2020-05-05 10:11:53 -04:00
trunc mon wholestage off 313 313 0 32.0 31.3 1.0X
trunc mon wholestage on 287 290 2 34.8 28.7 1.1X
[SPARK-29320][TESTS] Compare `sql/core` module in JDK8/11 (Part 1) ### What changes were proposed in this pull request? This PR regenerates the `sql/core` benchmarks in JDK8/11 to compare the result. In general, we compare the ratio instead of the time. However, in this PR, the average time is compared. This PR should be considered as a rough comparison. **A. EXPECTED CASES(JDK11 is faster in general)** - [x] BloomFilterBenchmark (JDK11 is faster except one case) - [x] BuiltInDataSourceWriteBenchmark (JDK11 is faster at CSV/ORC) - [x] CSVBenchmark (JDK11 is faster except five cases) - [x] ColumnarBatchBenchmark (JDK11 is faster at `boolean`/`string` and some cases in `int`/`array`) - [x] DatasetBenchmark (JDK11 is faster with `string`, but is slower for `long` type) - [x] ExternalAppendOnlyUnsafeRowArrayBenchmark (JDK11 is faster except two cases) - [x] ExtractBenchmark (JDK11 is faster except HOUR/MINUTE/SECOND/MILLISECONDS/MICROSECONDS) - [x] HashedRelationMetricsBenchmark (JDK11 is faster) - [x] JSONBenchmark (JDK11 is much faster except eight cases) - [x] JoinBenchmark (JDK11 is faster except five cases) - [x] OrcNestedSchemaPruningBenchmark (JDK11 is faster in nine cases) - [x] PrimitiveArrayBenchmark (JDK11 is faster) - [x] SortBenchmark (JDK11 is faster except `Arrays.sort` case) - [x] UDFBenchmark (N/A, values are too small) - [x] UnsafeArrayDataBenchmark (JDK11 is faster except one case) - [x] WideTableBenchmark (JDK11 is faster except two cases) **B. CASES WE NEED TO INVESTIGATE MORE LATER** - [x] AggregateBenchmark (JDK11 is slower in general) - [x] CompressionSchemeBenchmark (JDK11 is slower in general except `string`) - [x] DataSourceReadBenchmark (JDK11 is slower in general) - [x] DateTimeBenchmark (JDK11 is slightly slower in general except `parsing`) - [x] MakeDateTimeBenchmark (JDK11 is slower except two cases) - [x] MiscBenchmark (JDK11 is slower except ten cases) - [x] OrcV2NestedSchemaPruningBenchmark (JDK11 is slower) - [x] ParquetNestedSchemaPruningBenchmark (JDK11 is slower except six cases) - [x] RangeBenchmark (JDK11 is slower except one case) `FilterPushdownBenchmark/InExpressionBenchmark/WideSchemaBenchmark` will be compared later because it took long timer. ### Why are the changes needed? According to the result, there are some difference between JDK8/JDK11. This will be a baseline for the future improvement and comparison. Also, as a reproducible environment, the following environment is used. - Instance: `r3.xlarge` - OS: `CentOS Linux release 7.5.1804 (Core)` - JDK: - `OpenJDK Runtime Environment (build 1.8.0_222-b10)` - `OpenJDK Runtime Environment 18.9 (build 11.0.4+11-LTS)` ### Does this PR introduce any user-facing change? No. ### How was this patch tested? This is a test-only PR. We need to run benchmark. Closes #26003 from dongjoon-hyun/SPARK-29320. Authored-by: Dongjoon Hyun <dhyun@apple.com> Signed-off-by: Dongjoon Hyun <dhyun@apple.com>
2019-10-03 11:58:25 -04:00
[SPARK-31527][SQL][TESTS][FOLLOWUP] Fix the number of rows in `DateTimeBenchmark` ### What changes were proposed in this pull request? - Changed to the number of rows in benchmark cases from 3 to the actual number `N`. - Regenerated benchmark results in the environment: | Item | Description | | ---- | ----| | Region | us-west-2 (Oregon) | | Instance | r3.xlarge | | AMI | ubuntu/images/hvm-ssd/ubuntu-bionic-18.04-amd64-server-20190722.1 (ami-06f2f779464715dc5) | | Java | OpenJDK 64-Bit Server VM 1.8.0_242 and OpenJDK 64-Bit Server VM 11.0.6+10 | ### Why are the changes needed? The changes are needed to have: - Correct benchmark results - Base line for other perf improvements that can be checked in the same environment. ### Does this PR introduce _any_ user-facing change? No ### How was this patch tested? By running the benchmark and checking its output. Closes #28440 from MaxGekk/SPARK-31527-DateTimeBenchmark-followup. Authored-by: Max Gekk <max.gekk@gmail.com> Signed-off-by: Takeshi Yamamuro <yamamuro@apache.org>
2020-05-03 20:39:50 -04:00
OpenJDK 64-Bit Server VM 11.0.7+10-post-Ubuntu-2ubuntu218.04 on Linux 4.15.0-1063-aws
Intel(R) Xeon(R) CPU E5-2670 v2 @ 2.50GHz
[SPARK-29320][TESTS] Compare `sql/core` module in JDK8/11 (Part 1) ### What changes were proposed in this pull request? This PR regenerates the `sql/core` benchmarks in JDK8/11 to compare the result. In general, we compare the ratio instead of the time. However, in this PR, the average time is compared. This PR should be considered as a rough comparison. **A. EXPECTED CASES(JDK11 is faster in general)** - [x] BloomFilterBenchmark (JDK11 is faster except one case) - [x] BuiltInDataSourceWriteBenchmark (JDK11 is faster at CSV/ORC) - [x] CSVBenchmark (JDK11 is faster except five cases) - [x] ColumnarBatchBenchmark (JDK11 is faster at `boolean`/`string` and some cases in `int`/`array`) - [x] DatasetBenchmark (JDK11 is faster with `string`, but is slower for `long` type) - [x] ExternalAppendOnlyUnsafeRowArrayBenchmark (JDK11 is faster except two cases) - [x] ExtractBenchmark (JDK11 is faster except HOUR/MINUTE/SECOND/MILLISECONDS/MICROSECONDS) - [x] HashedRelationMetricsBenchmark (JDK11 is faster) - [x] JSONBenchmark (JDK11 is much faster except eight cases) - [x] JoinBenchmark (JDK11 is faster except five cases) - [x] OrcNestedSchemaPruningBenchmark (JDK11 is faster in nine cases) - [x] PrimitiveArrayBenchmark (JDK11 is faster) - [x] SortBenchmark (JDK11 is faster except `Arrays.sort` case) - [x] UDFBenchmark (N/A, values are too small) - [x] UnsafeArrayDataBenchmark (JDK11 is faster except one case) - [x] WideTableBenchmark (JDK11 is faster except two cases) **B. CASES WE NEED TO INVESTIGATE MORE LATER** - [x] AggregateBenchmark (JDK11 is slower in general) - [x] CompressionSchemeBenchmark (JDK11 is slower in general except `string`) - [x] DataSourceReadBenchmark (JDK11 is slower in general) - [x] DateTimeBenchmark (JDK11 is slightly slower in general except `parsing`) - [x] MakeDateTimeBenchmark (JDK11 is slower except two cases) - [x] MiscBenchmark (JDK11 is slower except ten cases) - [x] OrcV2NestedSchemaPruningBenchmark (JDK11 is slower) - [x] ParquetNestedSchemaPruningBenchmark (JDK11 is slower except six cases) - [x] RangeBenchmark (JDK11 is slower except one case) `FilterPushdownBenchmark/InExpressionBenchmark/WideSchemaBenchmark` will be compared later because it took long timer. ### Why are the changes needed? According to the result, there are some difference between JDK8/JDK11. This will be a baseline for the future improvement and comparison. Also, as a reproducible environment, the following environment is used. - Instance: `r3.xlarge` - OS: `CentOS Linux release 7.5.1804 (Core)` - JDK: - `OpenJDK Runtime Environment (build 1.8.0_222-b10)` - `OpenJDK Runtime Environment 18.9 (build 11.0.4+11-LTS)` ### Does this PR introduce any user-facing change? No. ### How was this patch tested? This is a test-only PR. We need to run benchmark. Closes #26003 from dongjoon-hyun/SPARK-29320. Authored-by: Dongjoon Hyun <dhyun@apple.com> Signed-off-by: Dongjoon Hyun <dhyun@apple.com>
2019-10-03 11:58:25 -04:00
trunc month: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative
------------------------------------------------------------------------------------------------------------------------
[SPARK-31630][SQL] Fix perf regression by skipping timestamps rebasing after some threshold ### What changes were proposed in this pull request? Skip timestamps rebasing after a global threshold when there is no difference between Julian and Gregorian calendars. This allows to avoid checking hash maps of switch points, and fixes perf regressions in `toJavaTimestamp()` and `fromJavaTimestamp()`. ### Why are the changes needed? The changes fix perf regressions of conversions to/from external type `java.sql.Timestamp`. Before (see the PR's results https://github.com/apache/spark/pull/28440): ``` ================================================================================================ Conversion from/to external types ================================================================================================ OpenJDK 64-Bit Server VM 1.8.0_252-8u252-b09-1~18.04-b09 on Linux 4.15.0-1063-aws Intel(R) Xeon(R) CPU E5-2670 v2 2.50GHz To/from Java's date-time: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative ------------------------------------------------------------------------------------------------------------------------ From java.sql.Timestamp 376 388 10 13.3 75.2 1.1X Collect java.sql.Timestamp 1878 1937 64 2.7 375.6 0.2X ``` After: ``` ================================================================================================ Conversion from/to external types ================================================================================================ OpenJDK 64-Bit Server VM 1.8.0_252-8u252-b09-1~18.04-b09 on Linux 4.15.0-1063-aws Intel(R) Xeon(R) CPU E5-2670 v2 2.50GHz To/from Java's date-time: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative ------------------------------------------------------------------------------------------------------------------------ From java.sql.Timestamp 249 264 24 20.1 49.8 1.7X Collect java.sql.Timestamp 1503 1523 24 3.3 300.5 0.3X ``` Perf improvements in average of: 1. From java.sql.Timestamp is ~ 34% 2. To java.sql.Timestamps is ~16% ### Does this PR introduce _any_ user-facing change? No ### How was this patch tested? By existing test suites `DateTimeUtilsSuite` and `RebaseDateTimeSuite`. Closes #28441 from MaxGekk/opt-rebase-common-threshold. Authored-by: Max Gekk <max.gekk@gmail.com> Signed-off-by: Wenchen Fan <wenchen@databricks.com>
2020-05-05 10:11:53 -04:00
trunc month wholestage off 310 310 0 32.3 31.0 1.0X
trunc month wholestage on 287 290 2 34.8 28.7 1.1X
[SPARK-29320][TESTS] Compare `sql/core` module in JDK8/11 (Part 1) ### What changes were proposed in this pull request? This PR regenerates the `sql/core` benchmarks in JDK8/11 to compare the result. In general, we compare the ratio instead of the time. However, in this PR, the average time is compared. This PR should be considered as a rough comparison. **A. EXPECTED CASES(JDK11 is faster in general)** - [x] BloomFilterBenchmark (JDK11 is faster except one case) - [x] BuiltInDataSourceWriteBenchmark (JDK11 is faster at CSV/ORC) - [x] CSVBenchmark (JDK11 is faster except five cases) - [x] ColumnarBatchBenchmark (JDK11 is faster at `boolean`/`string` and some cases in `int`/`array`) - [x] DatasetBenchmark (JDK11 is faster with `string`, but is slower for `long` type) - [x] ExternalAppendOnlyUnsafeRowArrayBenchmark (JDK11 is faster except two cases) - [x] ExtractBenchmark (JDK11 is faster except HOUR/MINUTE/SECOND/MILLISECONDS/MICROSECONDS) - [x] HashedRelationMetricsBenchmark (JDK11 is faster) - [x] JSONBenchmark (JDK11 is much faster except eight cases) - [x] JoinBenchmark (JDK11 is faster except five cases) - [x] OrcNestedSchemaPruningBenchmark (JDK11 is faster in nine cases) - [x] PrimitiveArrayBenchmark (JDK11 is faster) - [x] SortBenchmark (JDK11 is faster except `Arrays.sort` case) - [x] UDFBenchmark (N/A, values are too small) - [x] UnsafeArrayDataBenchmark (JDK11 is faster except one case) - [x] WideTableBenchmark (JDK11 is faster except two cases) **B. CASES WE NEED TO INVESTIGATE MORE LATER** - [x] AggregateBenchmark (JDK11 is slower in general) - [x] CompressionSchemeBenchmark (JDK11 is slower in general except `string`) - [x] DataSourceReadBenchmark (JDK11 is slower in general) - [x] DateTimeBenchmark (JDK11 is slightly slower in general except `parsing`) - [x] MakeDateTimeBenchmark (JDK11 is slower except two cases) - [x] MiscBenchmark (JDK11 is slower except ten cases) - [x] OrcV2NestedSchemaPruningBenchmark (JDK11 is slower) - [x] ParquetNestedSchemaPruningBenchmark (JDK11 is slower except six cases) - [x] RangeBenchmark (JDK11 is slower except one case) `FilterPushdownBenchmark/InExpressionBenchmark/WideSchemaBenchmark` will be compared later because it took long timer. ### Why are the changes needed? According to the result, there are some difference between JDK8/JDK11. This will be a baseline for the future improvement and comparison. Also, as a reproducible environment, the following environment is used. - Instance: `r3.xlarge` - OS: `CentOS Linux release 7.5.1804 (Core)` - JDK: - `OpenJDK Runtime Environment (build 1.8.0_222-b10)` - `OpenJDK Runtime Environment 18.9 (build 11.0.4+11-LTS)` ### Does this PR introduce any user-facing change? No. ### How was this patch tested? This is a test-only PR. We need to run benchmark. Closes #26003 from dongjoon-hyun/SPARK-29320. Authored-by: Dongjoon Hyun <dhyun@apple.com> Signed-off-by: Dongjoon Hyun <dhyun@apple.com>
2019-10-03 11:58:25 -04:00
[SPARK-31527][SQL][TESTS][FOLLOWUP] Fix the number of rows in `DateTimeBenchmark` ### What changes were proposed in this pull request? - Changed to the number of rows in benchmark cases from 3 to the actual number `N`. - Regenerated benchmark results in the environment: | Item | Description | | ---- | ----| | Region | us-west-2 (Oregon) | | Instance | r3.xlarge | | AMI | ubuntu/images/hvm-ssd/ubuntu-bionic-18.04-amd64-server-20190722.1 (ami-06f2f779464715dc5) | | Java | OpenJDK 64-Bit Server VM 1.8.0_242 and OpenJDK 64-Bit Server VM 11.0.6+10 | ### Why are the changes needed? The changes are needed to have: - Correct benchmark results - Base line for other perf improvements that can be checked in the same environment. ### Does this PR introduce _any_ user-facing change? No ### How was this patch tested? By running the benchmark and checking its output. Closes #28440 from MaxGekk/SPARK-31527-DateTimeBenchmark-followup. Authored-by: Max Gekk <max.gekk@gmail.com> Signed-off-by: Takeshi Yamamuro <yamamuro@apache.org>
2020-05-03 20:39:50 -04:00
OpenJDK 64-Bit Server VM 11.0.7+10-post-Ubuntu-2ubuntu218.04 on Linux 4.15.0-1063-aws
Intel(R) Xeon(R) CPU E5-2670 v2 @ 2.50GHz
[SPARK-29320][TESTS] Compare `sql/core` module in JDK8/11 (Part 1) ### What changes were proposed in this pull request? This PR regenerates the `sql/core` benchmarks in JDK8/11 to compare the result. In general, we compare the ratio instead of the time. However, in this PR, the average time is compared. This PR should be considered as a rough comparison. **A. EXPECTED CASES(JDK11 is faster in general)** - [x] BloomFilterBenchmark (JDK11 is faster except one case) - [x] BuiltInDataSourceWriteBenchmark (JDK11 is faster at CSV/ORC) - [x] CSVBenchmark (JDK11 is faster except five cases) - [x] ColumnarBatchBenchmark (JDK11 is faster at `boolean`/`string` and some cases in `int`/`array`) - [x] DatasetBenchmark (JDK11 is faster with `string`, but is slower for `long` type) - [x] ExternalAppendOnlyUnsafeRowArrayBenchmark (JDK11 is faster except two cases) - [x] ExtractBenchmark (JDK11 is faster except HOUR/MINUTE/SECOND/MILLISECONDS/MICROSECONDS) - [x] HashedRelationMetricsBenchmark (JDK11 is faster) - [x] JSONBenchmark (JDK11 is much faster except eight cases) - [x] JoinBenchmark (JDK11 is faster except five cases) - [x] OrcNestedSchemaPruningBenchmark (JDK11 is faster in nine cases) - [x] PrimitiveArrayBenchmark (JDK11 is faster) - [x] SortBenchmark (JDK11 is faster except `Arrays.sort` case) - [x] UDFBenchmark (N/A, values are too small) - [x] UnsafeArrayDataBenchmark (JDK11 is faster except one case) - [x] WideTableBenchmark (JDK11 is faster except two cases) **B. CASES WE NEED TO INVESTIGATE MORE LATER** - [x] AggregateBenchmark (JDK11 is slower in general) - [x] CompressionSchemeBenchmark (JDK11 is slower in general except `string`) - [x] DataSourceReadBenchmark (JDK11 is slower in general) - [x] DateTimeBenchmark (JDK11 is slightly slower in general except `parsing`) - [x] MakeDateTimeBenchmark (JDK11 is slower except two cases) - [x] MiscBenchmark (JDK11 is slower except ten cases) - [x] OrcV2NestedSchemaPruningBenchmark (JDK11 is slower) - [x] ParquetNestedSchemaPruningBenchmark (JDK11 is slower except six cases) - [x] RangeBenchmark (JDK11 is slower except one case) `FilterPushdownBenchmark/InExpressionBenchmark/WideSchemaBenchmark` will be compared later because it took long timer. ### Why are the changes needed? According to the result, there are some difference between JDK8/JDK11. This will be a baseline for the future improvement and comparison. Also, as a reproducible environment, the following environment is used. - Instance: `r3.xlarge` - OS: `CentOS Linux release 7.5.1804 (Core)` - JDK: - `OpenJDK Runtime Environment (build 1.8.0_222-b10)` - `OpenJDK Runtime Environment 18.9 (build 11.0.4+11-LTS)` ### Does this PR introduce any user-facing change? No. ### How was this patch tested? This is a test-only PR. We need to run benchmark. Closes #26003 from dongjoon-hyun/SPARK-29320. Authored-by: Dongjoon Hyun <dhyun@apple.com> Signed-off-by: Dongjoon Hyun <dhyun@apple.com>
2019-10-03 11:58:25 -04:00
trunc mm: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative
------------------------------------------------------------------------------------------------------------------------
[SPARK-31630][SQL] Fix perf regression by skipping timestamps rebasing after some threshold ### What changes were proposed in this pull request? Skip timestamps rebasing after a global threshold when there is no difference between Julian and Gregorian calendars. This allows to avoid checking hash maps of switch points, and fixes perf regressions in `toJavaTimestamp()` and `fromJavaTimestamp()`. ### Why are the changes needed? The changes fix perf regressions of conversions to/from external type `java.sql.Timestamp`. Before (see the PR's results https://github.com/apache/spark/pull/28440): ``` ================================================================================================ Conversion from/to external types ================================================================================================ OpenJDK 64-Bit Server VM 1.8.0_252-8u252-b09-1~18.04-b09 on Linux 4.15.0-1063-aws Intel(R) Xeon(R) CPU E5-2670 v2 2.50GHz To/from Java's date-time: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative ------------------------------------------------------------------------------------------------------------------------ From java.sql.Timestamp 376 388 10 13.3 75.2 1.1X Collect java.sql.Timestamp 1878 1937 64 2.7 375.6 0.2X ``` After: ``` ================================================================================================ Conversion from/to external types ================================================================================================ OpenJDK 64-Bit Server VM 1.8.0_252-8u252-b09-1~18.04-b09 on Linux 4.15.0-1063-aws Intel(R) Xeon(R) CPU E5-2670 v2 2.50GHz To/from Java's date-time: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative ------------------------------------------------------------------------------------------------------------------------ From java.sql.Timestamp 249 264 24 20.1 49.8 1.7X Collect java.sql.Timestamp 1503 1523 24 3.3 300.5 0.3X ``` Perf improvements in average of: 1. From java.sql.Timestamp is ~ 34% 2. To java.sql.Timestamps is ~16% ### Does this PR introduce _any_ user-facing change? No ### How was this patch tested? By existing test suites `DateTimeUtilsSuite` and `RebaseDateTimeSuite`. Closes #28441 from MaxGekk/opt-rebase-common-threshold. Authored-by: Max Gekk <max.gekk@gmail.com> Signed-off-by: Wenchen Fan <wenchen@databricks.com>
2020-05-05 10:11:53 -04:00
trunc mm wholestage off 311 312 1 32.1 31.1 1.0X
trunc mm wholestage on 287 296 9 34.8 28.7 1.1X
[SPARK-29320][TESTS] Compare `sql/core` module in JDK8/11 (Part 1) ### What changes were proposed in this pull request? This PR regenerates the `sql/core` benchmarks in JDK8/11 to compare the result. In general, we compare the ratio instead of the time. However, in this PR, the average time is compared. This PR should be considered as a rough comparison. **A. EXPECTED CASES(JDK11 is faster in general)** - [x] BloomFilterBenchmark (JDK11 is faster except one case) - [x] BuiltInDataSourceWriteBenchmark (JDK11 is faster at CSV/ORC) - [x] CSVBenchmark (JDK11 is faster except five cases) - [x] ColumnarBatchBenchmark (JDK11 is faster at `boolean`/`string` and some cases in `int`/`array`) - [x] DatasetBenchmark (JDK11 is faster with `string`, but is slower for `long` type) - [x] ExternalAppendOnlyUnsafeRowArrayBenchmark (JDK11 is faster except two cases) - [x] ExtractBenchmark (JDK11 is faster except HOUR/MINUTE/SECOND/MILLISECONDS/MICROSECONDS) - [x] HashedRelationMetricsBenchmark (JDK11 is faster) - [x] JSONBenchmark (JDK11 is much faster except eight cases) - [x] JoinBenchmark (JDK11 is faster except five cases) - [x] OrcNestedSchemaPruningBenchmark (JDK11 is faster in nine cases) - [x] PrimitiveArrayBenchmark (JDK11 is faster) - [x] SortBenchmark (JDK11 is faster except `Arrays.sort` case) - [x] UDFBenchmark (N/A, values are too small) - [x] UnsafeArrayDataBenchmark (JDK11 is faster except one case) - [x] WideTableBenchmark (JDK11 is faster except two cases) **B. CASES WE NEED TO INVESTIGATE MORE LATER** - [x] AggregateBenchmark (JDK11 is slower in general) - [x] CompressionSchemeBenchmark (JDK11 is slower in general except `string`) - [x] DataSourceReadBenchmark (JDK11 is slower in general) - [x] DateTimeBenchmark (JDK11 is slightly slower in general except `parsing`) - [x] MakeDateTimeBenchmark (JDK11 is slower except two cases) - [x] MiscBenchmark (JDK11 is slower except ten cases) - [x] OrcV2NestedSchemaPruningBenchmark (JDK11 is slower) - [x] ParquetNestedSchemaPruningBenchmark (JDK11 is slower except six cases) - [x] RangeBenchmark (JDK11 is slower except one case) `FilterPushdownBenchmark/InExpressionBenchmark/WideSchemaBenchmark` will be compared later because it took long timer. ### Why are the changes needed? According to the result, there are some difference between JDK8/JDK11. This will be a baseline for the future improvement and comparison. Also, as a reproducible environment, the following environment is used. - Instance: `r3.xlarge` - OS: `CentOS Linux release 7.5.1804 (Core)` - JDK: - `OpenJDK Runtime Environment (build 1.8.0_222-b10)` - `OpenJDK Runtime Environment 18.9 (build 11.0.4+11-LTS)` ### Does this PR introduce any user-facing change? No. ### How was this patch tested? This is a test-only PR. We need to run benchmark. Closes #26003 from dongjoon-hyun/SPARK-29320. Authored-by: Dongjoon Hyun <dhyun@apple.com> Signed-off-by: Dongjoon Hyun <dhyun@apple.com>
2019-10-03 11:58:25 -04:00
================================================================================================
Parsing
================================================================================================
[SPARK-31527][SQL][TESTS][FOLLOWUP] Fix the number of rows in `DateTimeBenchmark` ### What changes were proposed in this pull request? - Changed to the number of rows in benchmark cases from 3 to the actual number `N`. - Regenerated benchmark results in the environment: | Item | Description | | ---- | ----| | Region | us-west-2 (Oregon) | | Instance | r3.xlarge | | AMI | ubuntu/images/hvm-ssd/ubuntu-bionic-18.04-amd64-server-20190722.1 (ami-06f2f779464715dc5) | | Java | OpenJDK 64-Bit Server VM 1.8.0_242 and OpenJDK 64-Bit Server VM 11.0.6+10 | ### Why are the changes needed? The changes are needed to have: - Correct benchmark results - Base line for other perf improvements that can be checked in the same environment. ### Does this PR introduce _any_ user-facing change? No ### How was this patch tested? By running the benchmark and checking its output. Closes #28440 from MaxGekk/SPARK-31527-DateTimeBenchmark-followup. Authored-by: Max Gekk <max.gekk@gmail.com> Signed-off-by: Takeshi Yamamuro <yamamuro@apache.org>
2020-05-03 20:39:50 -04:00
OpenJDK 64-Bit Server VM 11.0.7+10-post-Ubuntu-2ubuntu218.04 on Linux 4.15.0-1063-aws
Intel(R) Xeon(R) CPU E5-2670 v2 @ 2.50GHz
[SPARK-29320][TESTS] Compare `sql/core` module in JDK8/11 (Part 1) ### What changes were proposed in this pull request? This PR regenerates the `sql/core` benchmarks in JDK8/11 to compare the result. In general, we compare the ratio instead of the time. However, in this PR, the average time is compared. This PR should be considered as a rough comparison. **A. EXPECTED CASES(JDK11 is faster in general)** - [x] BloomFilterBenchmark (JDK11 is faster except one case) - [x] BuiltInDataSourceWriteBenchmark (JDK11 is faster at CSV/ORC) - [x] CSVBenchmark (JDK11 is faster except five cases) - [x] ColumnarBatchBenchmark (JDK11 is faster at `boolean`/`string` and some cases in `int`/`array`) - [x] DatasetBenchmark (JDK11 is faster with `string`, but is slower for `long` type) - [x] ExternalAppendOnlyUnsafeRowArrayBenchmark (JDK11 is faster except two cases) - [x] ExtractBenchmark (JDK11 is faster except HOUR/MINUTE/SECOND/MILLISECONDS/MICROSECONDS) - [x] HashedRelationMetricsBenchmark (JDK11 is faster) - [x] JSONBenchmark (JDK11 is much faster except eight cases) - [x] JoinBenchmark (JDK11 is faster except five cases) - [x] OrcNestedSchemaPruningBenchmark (JDK11 is faster in nine cases) - [x] PrimitiveArrayBenchmark (JDK11 is faster) - [x] SortBenchmark (JDK11 is faster except `Arrays.sort` case) - [x] UDFBenchmark (N/A, values are too small) - [x] UnsafeArrayDataBenchmark (JDK11 is faster except one case) - [x] WideTableBenchmark (JDK11 is faster except two cases) **B. CASES WE NEED TO INVESTIGATE MORE LATER** - [x] AggregateBenchmark (JDK11 is slower in general) - [x] CompressionSchemeBenchmark (JDK11 is slower in general except `string`) - [x] DataSourceReadBenchmark (JDK11 is slower in general) - [x] DateTimeBenchmark (JDK11 is slightly slower in general except `parsing`) - [x] MakeDateTimeBenchmark (JDK11 is slower except two cases) - [x] MiscBenchmark (JDK11 is slower except ten cases) - [x] OrcV2NestedSchemaPruningBenchmark (JDK11 is slower) - [x] ParquetNestedSchemaPruningBenchmark (JDK11 is slower except six cases) - [x] RangeBenchmark (JDK11 is slower except one case) `FilterPushdownBenchmark/InExpressionBenchmark/WideSchemaBenchmark` will be compared later because it took long timer. ### Why are the changes needed? According to the result, there are some difference between JDK8/JDK11. This will be a baseline for the future improvement and comparison. Also, as a reproducible environment, the following environment is used. - Instance: `r3.xlarge` - OS: `CentOS Linux release 7.5.1804 (Core)` - JDK: - `OpenJDK Runtime Environment (build 1.8.0_222-b10)` - `OpenJDK Runtime Environment 18.9 (build 11.0.4+11-LTS)` ### Does this PR introduce any user-facing change? No. ### How was this patch tested? This is a test-only PR. We need to run benchmark. Closes #26003 from dongjoon-hyun/SPARK-29320. Authored-by: Dongjoon Hyun <dhyun@apple.com> Signed-off-by: Dongjoon Hyun <dhyun@apple.com>
2019-10-03 11:58:25 -04:00
to timestamp str: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative
------------------------------------------------------------------------------------------------------------------------
[SPARK-31630][SQL] Fix perf regression by skipping timestamps rebasing after some threshold ### What changes were proposed in this pull request? Skip timestamps rebasing after a global threshold when there is no difference between Julian and Gregorian calendars. This allows to avoid checking hash maps of switch points, and fixes perf regressions in `toJavaTimestamp()` and `fromJavaTimestamp()`. ### Why are the changes needed? The changes fix perf regressions of conversions to/from external type `java.sql.Timestamp`. Before (see the PR's results https://github.com/apache/spark/pull/28440): ``` ================================================================================================ Conversion from/to external types ================================================================================================ OpenJDK 64-Bit Server VM 1.8.0_252-8u252-b09-1~18.04-b09 on Linux 4.15.0-1063-aws Intel(R) Xeon(R) CPU E5-2670 v2 2.50GHz To/from Java's date-time: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative ------------------------------------------------------------------------------------------------------------------------ From java.sql.Timestamp 376 388 10 13.3 75.2 1.1X Collect java.sql.Timestamp 1878 1937 64 2.7 375.6 0.2X ``` After: ``` ================================================================================================ Conversion from/to external types ================================================================================================ OpenJDK 64-Bit Server VM 1.8.0_252-8u252-b09-1~18.04-b09 on Linux 4.15.0-1063-aws Intel(R) Xeon(R) CPU E5-2670 v2 2.50GHz To/from Java's date-time: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative ------------------------------------------------------------------------------------------------------------------------ From java.sql.Timestamp 249 264 24 20.1 49.8 1.7X Collect java.sql.Timestamp 1503 1523 24 3.3 300.5 0.3X ``` Perf improvements in average of: 1. From java.sql.Timestamp is ~ 34% 2. To java.sql.Timestamps is ~16% ### Does this PR introduce _any_ user-facing change? No ### How was this patch tested? By existing test suites `DateTimeUtilsSuite` and `RebaseDateTimeSuite`. Closes #28441 from MaxGekk/opt-rebase-common-threshold. Authored-by: Max Gekk <max.gekk@gmail.com> Signed-off-by: Wenchen Fan <wenchen@databricks.com>
2020-05-05 10:11:53 -04:00
to timestamp str wholestage off 169 170 1 5.9 168.9 1.0X
to timestamp str wholestage on 161 168 11 6.2 161.0 1.0X
[SPARK-29320][TESTS] Compare `sql/core` module in JDK8/11 (Part 1) ### What changes were proposed in this pull request? This PR regenerates the `sql/core` benchmarks in JDK8/11 to compare the result. In general, we compare the ratio instead of the time. However, in this PR, the average time is compared. This PR should be considered as a rough comparison. **A. EXPECTED CASES(JDK11 is faster in general)** - [x] BloomFilterBenchmark (JDK11 is faster except one case) - [x] BuiltInDataSourceWriteBenchmark (JDK11 is faster at CSV/ORC) - [x] CSVBenchmark (JDK11 is faster except five cases) - [x] ColumnarBatchBenchmark (JDK11 is faster at `boolean`/`string` and some cases in `int`/`array`) - [x] DatasetBenchmark (JDK11 is faster with `string`, but is slower for `long` type) - [x] ExternalAppendOnlyUnsafeRowArrayBenchmark (JDK11 is faster except two cases) - [x] ExtractBenchmark (JDK11 is faster except HOUR/MINUTE/SECOND/MILLISECONDS/MICROSECONDS) - [x] HashedRelationMetricsBenchmark (JDK11 is faster) - [x] JSONBenchmark (JDK11 is much faster except eight cases) - [x] JoinBenchmark (JDK11 is faster except five cases) - [x] OrcNestedSchemaPruningBenchmark (JDK11 is faster in nine cases) - [x] PrimitiveArrayBenchmark (JDK11 is faster) - [x] SortBenchmark (JDK11 is faster except `Arrays.sort` case) - [x] UDFBenchmark (N/A, values are too small) - [x] UnsafeArrayDataBenchmark (JDK11 is faster except one case) - [x] WideTableBenchmark (JDK11 is faster except two cases) **B. CASES WE NEED TO INVESTIGATE MORE LATER** - [x] AggregateBenchmark (JDK11 is slower in general) - [x] CompressionSchemeBenchmark (JDK11 is slower in general except `string`) - [x] DataSourceReadBenchmark (JDK11 is slower in general) - [x] DateTimeBenchmark (JDK11 is slightly slower in general except `parsing`) - [x] MakeDateTimeBenchmark (JDK11 is slower except two cases) - [x] MiscBenchmark (JDK11 is slower except ten cases) - [x] OrcV2NestedSchemaPruningBenchmark (JDK11 is slower) - [x] ParquetNestedSchemaPruningBenchmark (JDK11 is slower except six cases) - [x] RangeBenchmark (JDK11 is slower except one case) `FilterPushdownBenchmark/InExpressionBenchmark/WideSchemaBenchmark` will be compared later because it took long timer. ### Why are the changes needed? According to the result, there are some difference between JDK8/JDK11. This will be a baseline for the future improvement and comparison. Also, as a reproducible environment, the following environment is used. - Instance: `r3.xlarge` - OS: `CentOS Linux release 7.5.1804 (Core)` - JDK: - `OpenJDK Runtime Environment (build 1.8.0_222-b10)` - `OpenJDK Runtime Environment 18.9 (build 11.0.4+11-LTS)` ### Does this PR introduce any user-facing change? No. ### How was this patch tested? This is a test-only PR. We need to run benchmark. Closes #26003 from dongjoon-hyun/SPARK-29320. Authored-by: Dongjoon Hyun <dhyun@apple.com> Signed-off-by: Dongjoon Hyun <dhyun@apple.com>
2019-10-03 11:58:25 -04:00
[SPARK-31527][SQL][TESTS][FOLLOWUP] Fix the number of rows in `DateTimeBenchmark` ### What changes were proposed in this pull request? - Changed to the number of rows in benchmark cases from 3 to the actual number `N`. - Regenerated benchmark results in the environment: | Item | Description | | ---- | ----| | Region | us-west-2 (Oregon) | | Instance | r3.xlarge | | AMI | ubuntu/images/hvm-ssd/ubuntu-bionic-18.04-amd64-server-20190722.1 (ami-06f2f779464715dc5) | | Java | OpenJDK 64-Bit Server VM 1.8.0_242 and OpenJDK 64-Bit Server VM 11.0.6+10 | ### Why are the changes needed? The changes are needed to have: - Correct benchmark results - Base line for other perf improvements that can be checked in the same environment. ### Does this PR introduce _any_ user-facing change? No ### How was this patch tested? By running the benchmark and checking its output. Closes #28440 from MaxGekk/SPARK-31527-DateTimeBenchmark-followup. Authored-by: Max Gekk <max.gekk@gmail.com> Signed-off-by: Takeshi Yamamuro <yamamuro@apache.org>
2020-05-03 20:39:50 -04:00
OpenJDK 64-Bit Server VM 11.0.7+10-post-Ubuntu-2ubuntu218.04 on Linux 4.15.0-1063-aws
Intel(R) Xeon(R) CPU E5-2670 v2 @ 2.50GHz
[SPARK-29320][TESTS] Compare `sql/core` module in JDK8/11 (Part 1) ### What changes were proposed in this pull request? This PR regenerates the `sql/core` benchmarks in JDK8/11 to compare the result. In general, we compare the ratio instead of the time. However, in this PR, the average time is compared. This PR should be considered as a rough comparison. **A. EXPECTED CASES(JDK11 is faster in general)** - [x] BloomFilterBenchmark (JDK11 is faster except one case) - [x] BuiltInDataSourceWriteBenchmark (JDK11 is faster at CSV/ORC) - [x] CSVBenchmark (JDK11 is faster except five cases) - [x] ColumnarBatchBenchmark (JDK11 is faster at `boolean`/`string` and some cases in `int`/`array`) - [x] DatasetBenchmark (JDK11 is faster with `string`, but is slower for `long` type) - [x] ExternalAppendOnlyUnsafeRowArrayBenchmark (JDK11 is faster except two cases) - [x] ExtractBenchmark (JDK11 is faster except HOUR/MINUTE/SECOND/MILLISECONDS/MICROSECONDS) - [x] HashedRelationMetricsBenchmark (JDK11 is faster) - [x] JSONBenchmark (JDK11 is much faster except eight cases) - [x] JoinBenchmark (JDK11 is faster except five cases) - [x] OrcNestedSchemaPruningBenchmark (JDK11 is faster in nine cases) - [x] PrimitiveArrayBenchmark (JDK11 is faster) - [x] SortBenchmark (JDK11 is faster except `Arrays.sort` case) - [x] UDFBenchmark (N/A, values are too small) - [x] UnsafeArrayDataBenchmark (JDK11 is faster except one case) - [x] WideTableBenchmark (JDK11 is faster except two cases) **B. CASES WE NEED TO INVESTIGATE MORE LATER** - [x] AggregateBenchmark (JDK11 is slower in general) - [x] CompressionSchemeBenchmark (JDK11 is slower in general except `string`) - [x] DataSourceReadBenchmark (JDK11 is slower in general) - [x] DateTimeBenchmark (JDK11 is slightly slower in general except `parsing`) - [x] MakeDateTimeBenchmark (JDK11 is slower except two cases) - [x] MiscBenchmark (JDK11 is slower except ten cases) - [x] OrcV2NestedSchemaPruningBenchmark (JDK11 is slower) - [x] ParquetNestedSchemaPruningBenchmark (JDK11 is slower except six cases) - [x] RangeBenchmark (JDK11 is slower except one case) `FilterPushdownBenchmark/InExpressionBenchmark/WideSchemaBenchmark` will be compared later because it took long timer. ### Why are the changes needed? According to the result, there are some difference between JDK8/JDK11. This will be a baseline for the future improvement and comparison. Also, as a reproducible environment, the following environment is used. - Instance: `r3.xlarge` - OS: `CentOS Linux release 7.5.1804 (Core)` - JDK: - `OpenJDK Runtime Environment (build 1.8.0_222-b10)` - `OpenJDK Runtime Environment 18.9 (build 11.0.4+11-LTS)` ### Does this PR introduce any user-facing change? No. ### How was this patch tested? This is a test-only PR. We need to run benchmark. Closes #26003 from dongjoon-hyun/SPARK-29320. Authored-by: Dongjoon Hyun <dhyun@apple.com> Signed-off-by: Dongjoon Hyun <dhyun@apple.com>
2019-10-03 11:58:25 -04:00
to_timestamp: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative
------------------------------------------------------------------------------------------------------------------------
[SPARK-31630][SQL] Fix perf regression by skipping timestamps rebasing after some threshold ### What changes were proposed in this pull request? Skip timestamps rebasing after a global threshold when there is no difference between Julian and Gregorian calendars. This allows to avoid checking hash maps of switch points, and fixes perf regressions in `toJavaTimestamp()` and `fromJavaTimestamp()`. ### Why are the changes needed? The changes fix perf regressions of conversions to/from external type `java.sql.Timestamp`. Before (see the PR's results https://github.com/apache/spark/pull/28440): ``` ================================================================================================ Conversion from/to external types ================================================================================================ OpenJDK 64-Bit Server VM 1.8.0_252-8u252-b09-1~18.04-b09 on Linux 4.15.0-1063-aws Intel(R) Xeon(R) CPU E5-2670 v2 2.50GHz To/from Java's date-time: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative ------------------------------------------------------------------------------------------------------------------------ From java.sql.Timestamp 376 388 10 13.3 75.2 1.1X Collect java.sql.Timestamp 1878 1937 64 2.7 375.6 0.2X ``` After: ``` ================================================================================================ Conversion from/to external types ================================================================================================ OpenJDK 64-Bit Server VM 1.8.0_252-8u252-b09-1~18.04-b09 on Linux 4.15.0-1063-aws Intel(R) Xeon(R) CPU E5-2670 v2 2.50GHz To/from Java's date-time: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative ------------------------------------------------------------------------------------------------------------------------ From java.sql.Timestamp 249 264 24 20.1 49.8 1.7X Collect java.sql.Timestamp 1503 1523 24 3.3 300.5 0.3X ``` Perf improvements in average of: 1. From java.sql.Timestamp is ~ 34% 2. To java.sql.Timestamps is ~16% ### Does this PR introduce _any_ user-facing change? No ### How was this patch tested? By existing test suites `DateTimeUtilsSuite` and `RebaseDateTimeSuite`. Closes #28441 from MaxGekk/opt-rebase-common-threshold. Authored-by: Max Gekk <max.gekk@gmail.com> Signed-off-by: Wenchen Fan <wenchen@databricks.com>
2020-05-05 10:11:53 -04:00
to_timestamp wholestage off 1360 1361 1 0.7 1359.6 1.0X
to_timestamp wholestage on 1362 1366 6 0.7 1362.0 1.0X
[SPARK-29320][TESTS] Compare `sql/core` module in JDK8/11 (Part 1) ### What changes were proposed in this pull request? This PR regenerates the `sql/core` benchmarks in JDK8/11 to compare the result. In general, we compare the ratio instead of the time. However, in this PR, the average time is compared. This PR should be considered as a rough comparison. **A. EXPECTED CASES(JDK11 is faster in general)** - [x] BloomFilterBenchmark (JDK11 is faster except one case) - [x] BuiltInDataSourceWriteBenchmark (JDK11 is faster at CSV/ORC) - [x] CSVBenchmark (JDK11 is faster except five cases) - [x] ColumnarBatchBenchmark (JDK11 is faster at `boolean`/`string` and some cases in `int`/`array`) - [x] DatasetBenchmark (JDK11 is faster with `string`, but is slower for `long` type) - [x] ExternalAppendOnlyUnsafeRowArrayBenchmark (JDK11 is faster except two cases) - [x] ExtractBenchmark (JDK11 is faster except HOUR/MINUTE/SECOND/MILLISECONDS/MICROSECONDS) - [x] HashedRelationMetricsBenchmark (JDK11 is faster) - [x] JSONBenchmark (JDK11 is much faster except eight cases) - [x] JoinBenchmark (JDK11 is faster except five cases) - [x] OrcNestedSchemaPruningBenchmark (JDK11 is faster in nine cases) - [x] PrimitiveArrayBenchmark (JDK11 is faster) - [x] SortBenchmark (JDK11 is faster except `Arrays.sort` case) - [x] UDFBenchmark (N/A, values are too small) - [x] UnsafeArrayDataBenchmark (JDK11 is faster except one case) - [x] WideTableBenchmark (JDK11 is faster except two cases) **B. CASES WE NEED TO INVESTIGATE MORE LATER** - [x] AggregateBenchmark (JDK11 is slower in general) - [x] CompressionSchemeBenchmark (JDK11 is slower in general except `string`) - [x] DataSourceReadBenchmark (JDK11 is slower in general) - [x] DateTimeBenchmark (JDK11 is slightly slower in general except `parsing`) - [x] MakeDateTimeBenchmark (JDK11 is slower except two cases) - [x] MiscBenchmark (JDK11 is slower except ten cases) - [x] OrcV2NestedSchemaPruningBenchmark (JDK11 is slower) - [x] ParquetNestedSchemaPruningBenchmark (JDK11 is slower except six cases) - [x] RangeBenchmark (JDK11 is slower except one case) `FilterPushdownBenchmark/InExpressionBenchmark/WideSchemaBenchmark` will be compared later because it took long timer. ### Why are the changes needed? According to the result, there are some difference between JDK8/JDK11. This will be a baseline for the future improvement and comparison. Also, as a reproducible environment, the following environment is used. - Instance: `r3.xlarge` - OS: `CentOS Linux release 7.5.1804 (Core)` - JDK: - `OpenJDK Runtime Environment (build 1.8.0_222-b10)` - `OpenJDK Runtime Environment 18.9 (build 11.0.4+11-LTS)` ### Does this PR introduce any user-facing change? No. ### How was this patch tested? This is a test-only PR. We need to run benchmark. Closes #26003 from dongjoon-hyun/SPARK-29320. Authored-by: Dongjoon Hyun <dhyun@apple.com> Signed-off-by: Dongjoon Hyun <dhyun@apple.com>
2019-10-03 11:58:25 -04:00
[SPARK-31527][SQL][TESTS][FOLLOWUP] Fix the number of rows in `DateTimeBenchmark` ### What changes were proposed in this pull request? - Changed to the number of rows in benchmark cases from 3 to the actual number `N`. - Regenerated benchmark results in the environment: | Item | Description | | ---- | ----| | Region | us-west-2 (Oregon) | | Instance | r3.xlarge | | AMI | ubuntu/images/hvm-ssd/ubuntu-bionic-18.04-amd64-server-20190722.1 (ami-06f2f779464715dc5) | | Java | OpenJDK 64-Bit Server VM 1.8.0_242 and OpenJDK 64-Bit Server VM 11.0.6+10 | ### Why are the changes needed? The changes are needed to have: - Correct benchmark results - Base line for other perf improvements that can be checked in the same environment. ### Does this PR introduce _any_ user-facing change? No ### How was this patch tested? By running the benchmark and checking its output. Closes #28440 from MaxGekk/SPARK-31527-DateTimeBenchmark-followup. Authored-by: Max Gekk <max.gekk@gmail.com> Signed-off-by: Takeshi Yamamuro <yamamuro@apache.org>
2020-05-03 20:39:50 -04:00
OpenJDK 64-Bit Server VM 11.0.7+10-post-Ubuntu-2ubuntu218.04 on Linux 4.15.0-1063-aws
Intel(R) Xeon(R) CPU E5-2670 v2 @ 2.50GHz
[SPARK-29320][TESTS] Compare `sql/core` module in JDK8/11 (Part 1) ### What changes were proposed in this pull request? This PR regenerates the `sql/core` benchmarks in JDK8/11 to compare the result. In general, we compare the ratio instead of the time. However, in this PR, the average time is compared. This PR should be considered as a rough comparison. **A. EXPECTED CASES(JDK11 is faster in general)** - [x] BloomFilterBenchmark (JDK11 is faster except one case) - [x] BuiltInDataSourceWriteBenchmark (JDK11 is faster at CSV/ORC) - [x] CSVBenchmark (JDK11 is faster except five cases) - [x] ColumnarBatchBenchmark (JDK11 is faster at `boolean`/`string` and some cases in `int`/`array`) - [x] DatasetBenchmark (JDK11 is faster with `string`, but is slower for `long` type) - [x] ExternalAppendOnlyUnsafeRowArrayBenchmark (JDK11 is faster except two cases) - [x] ExtractBenchmark (JDK11 is faster except HOUR/MINUTE/SECOND/MILLISECONDS/MICROSECONDS) - [x] HashedRelationMetricsBenchmark (JDK11 is faster) - [x] JSONBenchmark (JDK11 is much faster except eight cases) - [x] JoinBenchmark (JDK11 is faster except five cases) - [x] OrcNestedSchemaPruningBenchmark (JDK11 is faster in nine cases) - [x] PrimitiveArrayBenchmark (JDK11 is faster) - [x] SortBenchmark (JDK11 is faster except `Arrays.sort` case) - [x] UDFBenchmark (N/A, values are too small) - [x] UnsafeArrayDataBenchmark (JDK11 is faster except one case) - [x] WideTableBenchmark (JDK11 is faster except two cases) **B. CASES WE NEED TO INVESTIGATE MORE LATER** - [x] AggregateBenchmark (JDK11 is slower in general) - [x] CompressionSchemeBenchmark (JDK11 is slower in general except `string`) - [x] DataSourceReadBenchmark (JDK11 is slower in general) - [x] DateTimeBenchmark (JDK11 is slightly slower in general except `parsing`) - [x] MakeDateTimeBenchmark (JDK11 is slower except two cases) - [x] MiscBenchmark (JDK11 is slower except ten cases) - [x] OrcV2NestedSchemaPruningBenchmark (JDK11 is slower) - [x] ParquetNestedSchemaPruningBenchmark (JDK11 is slower except six cases) - [x] RangeBenchmark (JDK11 is slower except one case) `FilterPushdownBenchmark/InExpressionBenchmark/WideSchemaBenchmark` will be compared later because it took long timer. ### Why are the changes needed? According to the result, there are some difference between JDK8/JDK11. This will be a baseline for the future improvement and comparison. Also, as a reproducible environment, the following environment is used. - Instance: `r3.xlarge` - OS: `CentOS Linux release 7.5.1804 (Core)` - JDK: - `OpenJDK Runtime Environment (build 1.8.0_222-b10)` - `OpenJDK Runtime Environment 18.9 (build 11.0.4+11-LTS)` ### Does this PR introduce any user-facing change? No. ### How was this patch tested? This is a test-only PR. We need to run benchmark. Closes #26003 from dongjoon-hyun/SPARK-29320. Authored-by: Dongjoon Hyun <dhyun@apple.com> Signed-off-by: Dongjoon Hyun <dhyun@apple.com>
2019-10-03 11:58:25 -04:00
to_unix_timestamp: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative
------------------------------------------------------------------------------------------------------------------------
[SPARK-31630][SQL] Fix perf regression by skipping timestamps rebasing after some threshold ### What changes were proposed in this pull request? Skip timestamps rebasing after a global threshold when there is no difference between Julian and Gregorian calendars. This allows to avoid checking hash maps of switch points, and fixes perf regressions in `toJavaTimestamp()` and `fromJavaTimestamp()`. ### Why are the changes needed? The changes fix perf regressions of conversions to/from external type `java.sql.Timestamp`. Before (see the PR's results https://github.com/apache/spark/pull/28440): ``` ================================================================================================ Conversion from/to external types ================================================================================================ OpenJDK 64-Bit Server VM 1.8.0_252-8u252-b09-1~18.04-b09 on Linux 4.15.0-1063-aws Intel(R) Xeon(R) CPU E5-2670 v2 2.50GHz To/from Java's date-time: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative ------------------------------------------------------------------------------------------------------------------------ From java.sql.Timestamp 376 388 10 13.3 75.2 1.1X Collect java.sql.Timestamp 1878 1937 64 2.7 375.6 0.2X ``` After: ``` ================================================================================================ Conversion from/to external types ================================================================================================ OpenJDK 64-Bit Server VM 1.8.0_252-8u252-b09-1~18.04-b09 on Linux 4.15.0-1063-aws Intel(R) Xeon(R) CPU E5-2670 v2 2.50GHz To/from Java's date-time: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative ------------------------------------------------------------------------------------------------------------------------ From java.sql.Timestamp 249 264 24 20.1 49.8 1.7X Collect java.sql.Timestamp 1503 1523 24 3.3 300.5 0.3X ``` Perf improvements in average of: 1. From java.sql.Timestamp is ~ 34% 2. To java.sql.Timestamps is ~16% ### Does this PR introduce _any_ user-facing change? No ### How was this patch tested? By existing test suites `DateTimeUtilsSuite` and `RebaseDateTimeSuite`. Closes #28441 from MaxGekk/opt-rebase-common-threshold. Authored-by: Max Gekk <max.gekk@gmail.com> Signed-off-by: Wenchen Fan <wenchen@databricks.com>
2020-05-05 10:11:53 -04:00
to_unix_timestamp wholestage off 1343 1346 4 0.7 1342.6 1.0X
to_unix_timestamp wholestage on 1356 1359 2 0.7 1356.2 1.0X
[SPARK-29320][TESTS] Compare `sql/core` module in JDK8/11 (Part 1) ### What changes were proposed in this pull request? This PR regenerates the `sql/core` benchmarks in JDK8/11 to compare the result. In general, we compare the ratio instead of the time. However, in this PR, the average time is compared. This PR should be considered as a rough comparison. **A. EXPECTED CASES(JDK11 is faster in general)** - [x] BloomFilterBenchmark (JDK11 is faster except one case) - [x] BuiltInDataSourceWriteBenchmark (JDK11 is faster at CSV/ORC) - [x] CSVBenchmark (JDK11 is faster except five cases) - [x] ColumnarBatchBenchmark (JDK11 is faster at `boolean`/`string` and some cases in `int`/`array`) - [x] DatasetBenchmark (JDK11 is faster with `string`, but is slower for `long` type) - [x] ExternalAppendOnlyUnsafeRowArrayBenchmark (JDK11 is faster except two cases) - [x] ExtractBenchmark (JDK11 is faster except HOUR/MINUTE/SECOND/MILLISECONDS/MICROSECONDS) - [x] HashedRelationMetricsBenchmark (JDK11 is faster) - [x] JSONBenchmark (JDK11 is much faster except eight cases) - [x] JoinBenchmark (JDK11 is faster except five cases) - [x] OrcNestedSchemaPruningBenchmark (JDK11 is faster in nine cases) - [x] PrimitiveArrayBenchmark (JDK11 is faster) - [x] SortBenchmark (JDK11 is faster except `Arrays.sort` case) - [x] UDFBenchmark (N/A, values are too small) - [x] UnsafeArrayDataBenchmark (JDK11 is faster except one case) - [x] WideTableBenchmark (JDK11 is faster except two cases) **B. CASES WE NEED TO INVESTIGATE MORE LATER** - [x] AggregateBenchmark (JDK11 is slower in general) - [x] CompressionSchemeBenchmark (JDK11 is slower in general except `string`) - [x] DataSourceReadBenchmark (JDK11 is slower in general) - [x] DateTimeBenchmark (JDK11 is slightly slower in general except `parsing`) - [x] MakeDateTimeBenchmark (JDK11 is slower except two cases) - [x] MiscBenchmark (JDK11 is slower except ten cases) - [x] OrcV2NestedSchemaPruningBenchmark (JDK11 is slower) - [x] ParquetNestedSchemaPruningBenchmark (JDK11 is slower except six cases) - [x] RangeBenchmark (JDK11 is slower except one case) `FilterPushdownBenchmark/InExpressionBenchmark/WideSchemaBenchmark` will be compared later because it took long timer. ### Why are the changes needed? According to the result, there are some difference between JDK8/JDK11. This will be a baseline for the future improvement and comparison. Also, as a reproducible environment, the following environment is used. - Instance: `r3.xlarge` - OS: `CentOS Linux release 7.5.1804 (Core)` - JDK: - `OpenJDK Runtime Environment (build 1.8.0_222-b10)` - `OpenJDK Runtime Environment 18.9 (build 11.0.4+11-LTS)` ### Does this PR introduce any user-facing change? No. ### How was this patch tested? This is a test-only PR. We need to run benchmark. Closes #26003 from dongjoon-hyun/SPARK-29320. Authored-by: Dongjoon Hyun <dhyun@apple.com> Signed-off-by: Dongjoon Hyun <dhyun@apple.com>
2019-10-03 11:58:25 -04:00
[SPARK-31527][SQL][TESTS][FOLLOWUP] Fix the number of rows in `DateTimeBenchmark` ### What changes were proposed in this pull request? - Changed to the number of rows in benchmark cases from 3 to the actual number `N`. - Regenerated benchmark results in the environment: | Item | Description | | ---- | ----| | Region | us-west-2 (Oregon) | | Instance | r3.xlarge | | AMI | ubuntu/images/hvm-ssd/ubuntu-bionic-18.04-amd64-server-20190722.1 (ami-06f2f779464715dc5) | | Java | OpenJDK 64-Bit Server VM 1.8.0_242 and OpenJDK 64-Bit Server VM 11.0.6+10 | ### Why are the changes needed? The changes are needed to have: - Correct benchmark results - Base line for other perf improvements that can be checked in the same environment. ### Does this PR introduce _any_ user-facing change? No ### How was this patch tested? By running the benchmark and checking its output. Closes #28440 from MaxGekk/SPARK-31527-DateTimeBenchmark-followup. Authored-by: Max Gekk <max.gekk@gmail.com> Signed-off-by: Takeshi Yamamuro <yamamuro@apache.org>
2020-05-03 20:39:50 -04:00
OpenJDK 64-Bit Server VM 11.0.7+10-post-Ubuntu-2ubuntu218.04 on Linux 4.15.0-1063-aws
Intel(R) Xeon(R) CPU E5-2670 v2 @ 2.50GHz
[SPARK-29320][TESTS] Compare `sql/core` module in JDK8/11 (Part 1) ### What changes were proposed in this pull request? This PR regenerates the `sql/core` benchmarks in JDK8/11 to compare the result. In general, we compare the ratio instead of the time. However, in this PR, the average time is compared. This PR should be considered as a rough comparison. **A. EXPECTED CASES(JDK11 is faster in general)** - [x] BloomFilterBenchmark (JDK11 is faster except one case) - [x] BuiltInDataSourceWriteBenchmark (JDK11 is faster at CSV/ORC) - [x] CSVBenchmark (JDK11 is faster except five cases) - [x] ColumnarBatchBenchmark (JDK11 is faster at `boolean`/`string` and some cases in `int`/`array`) - [x] DatasetBenchmark (JDK11 is faster with `string`, but is slower for `long` type) - [x] ExternalAppendOnlyUnsafeRowArrayBenchmark (JDK11 is faster except two cases) - [x] ExtractBenchmark (JDK11 is faster except HOUR/MINUTE/SECOND/MILLISECONDS/MICROSECONDS) - [x] HashedRelationMetricsBenchmark (JDK11 is faster) - [x] JSONBenchmark (JDK11 is much faster except eight cases) - [x] JoinBenchmark (JDK11 is faster except five cases) - [x] OrcNestedSchemaPruningBenchmark (JDK11 is faster in nine cases) - [x] PrimitiveArrayBenchmark (JDK11 is faster) - [x] SortBenchmark (JDK11 is faster except `Arrays.sort` case) - [x] UDFBenchmark (N/A, values are too small) - [x] UnsafeArrayDataBenchmark (JDK11 is faster except one case) - [x] WideTableBenchmark (JDK11 is faster except two cases) **B. CASES WE NEED TO INVESTIGATE MORE LATER** - [x] AggregateBenchmark (JDK11 is slower in general) - [x] CompressionSchemeBenchmark (JDK11 is slower in general except `string`) - [x] DataSourceReadBenchmark (JDK11 is slower in general) - [x] DateTimeBenchmark (JDK11 is slightly slower in general except `parsing`) - [x] MakeDateTimeBenchmark (JDK11 is slower except two cases) - [x] MiscBenchmark (JDK11 is slower except ten cases) - [x] OrcV2NestedSchemaPruningBenchmark (JDK11 is slower) - [x] ParquetNestedSchemaPruningBenchmark (JDK11 is slower except six cases) - [x] RangeBenchmark (JDK11 is slower except one case) `FilterPushdownBenchmark/InExpressionBenchmark/WideSchemaBenchmark` will be compared later because it took long timer. ### Why are the changes needed? According to the result, there are some difference between JDK8/JDK11. This will be a baseline for the future improvement and comparison. Also, as a reproducible environment, the following environment is used. - Instance: `r3.xlarge` - OS: `CentOS Linux release 7.5.1804 (Core)` - JDK: - `OpenJDK Runtime Environment (build 1.8.0_222-b10)` - `OpenJDK Runtime Environment 18.9 (build 11.0.4+11-LTS)` ### Does this PR introduce any user-facing change? No. ### How was this patch tested? This is a test-only PR. We need to run benchmark. Closes #26003 from dongjoon-hyun/SPARK-29320. Authored-by: Dongjoon Hyun <dhyun@apple.com> Signed-off-by: Dongjoon Hyun <dhyun@apple.com>
2019-10-03 11:58:25 -04:00
to date str: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative
------------------------------------------------------------------------------------------------------------------------
[SPARK-31630][SQL] Fix perf regression by skipping timestamps rebasing after some threshold ### What changes were proposed in this pull request? Skip timestamps rebasing after a global threshold when there is no difference between Julian and Gregorian calendars. This allows to avoid checking hash maps of switch points, and fixes perf regressions in `toJavaTimestamp()` and `fromJavaTimestamp()`. ### Why are the changes needed? The changes fix perf regressions of conversions to/from external type `java.sql.Timestamp`. Before (see the PR's results https://github.com/apache/spark/pull/28440): ``` ================================================================================================ Conversion from/to external types ================================================================================================ OpenJDK 64-Bit Server VM 1.8.0_252-8u252-b09-1~18.04-b09 on Linux 4.15.0-1063-aws Intel(R) Xeon(R) CPU E5-2670 v2 2.50GHz To/from Java's date-time: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative ------------------------------------------------------------------------------------------------------------------------ From java.sql.Timestamp 376 388 10 13.3 75.2 1.1X Collect java.sql.Timestamp 1878 1937 64 2.7 375.6 0.2X ``` After: ``` ================================================================================================ Conversion from/to external types ================================================================================================ OpenJDK 64-Bit Server VM 1.8.0_252-8u252-b09-1~18.04-b09 on Linux 4.15.0-1063-aws Intel(R) Xeon(R) CPU E5-2670 v2 2.50GHz To/from Java's date-time: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative ------------------------------------------------------------------------------------------------------------------------ From java.sql.Timestamp 249 264 24 20.1 49.8 1.7X Collect java.sql.Timestamp 1503 1523 24 3.3 300.5 0.3X ``` Perf improvements in average of: 1. From java.sql.Timestamp is ~ 34% 2. To java.sql.Timestamps is ~16% ### Does this PR introduce _any_ user-facing change? No ### How was this patch tested? By existing test suites `DateTimeUtilsSuite` and `RebaseDateTimeSuite`. Closes #28441 from MaxGekk/opt-rebase-common-threshold. Authored-by: Max Gekk <max.gekk@gmail.com> Signed-off-by: Wenchen Fan <wenchen@databricks.com>
2020-05-05 10:11:53 -04:00
to date str wholestage off 227 230 4 4.4 227.0 1.0X
to date str wholestage on 299 302 3 3.3 299.0 0.8X
[SPARK-29320][TESTS] Compare `sql/core` module in JDK8/11 (Part 1) ### What changes were proposed in this pull request? This PR regenerates the `sql/core` benchmarks in JDK8/11 to compare the result. In general, we compare the ratio instead of the time. However, in this PR, the average time is compared. This PR should be considered as a rough comparison. **A. EXPECTED CASES(JDK11 is faster in general)** - [x] BloomFilterBenchmark (JDK11 is faster except one case) - [x] BuiltInDataSourceWriteBenchmark (JDK11 is faster at CSV/ORC) - [x] CSVBenchmark (JDK11 is faster except five cases) - [x] ColumnarBatchBenchmark (JDK11 is faster at `boolean`/`string` and some cases in `int`/`array`) - [x] DatasetBenchmark (JDK11 is faster with `string`, but is slower for `long` type) - [x] ExternalAppendOnlyUnsafeRowArrayBenchmark (JDK11 is faster except two cases) - [x] ExtractBenchmark (JDK11 is faster except HOUR/MINUTE/SECOND/MILLISECONDS/MICROSECONDS) - [x] HashedRelationMetricsBenchmark (JDK11 is faster) - [x] JSONBenchmark (JDK11 is much faster except eight cases) - [x] JoinBenchmark (JDK11 is faster except five cases) - [x] OrcNestedSchemaPruningBenchmark (JDK11 is faster in nine cases) - [x] PrimitiveArrayBenchmark (JDK11 is faster) - [x] SortBenchmark (JDK11 is faster except `Arrays.sort` case) - [x] UDFBenchmark (N/A, values are too small) - [x] UnsafeArrayDataBenchmark (JDK11 is faster except one case) - [x] WideTableBenchmark (JDK11 is faster except two cases) **B. CASES WE NEED TO INVESTIGATE MORE LATER** - [x] AggregateBenchmark (JDK11 is slower in general) - [x] CompressionSchemeBenchmark (JDK11 is slower in general except `string`) - [x] DataSourceReadBenchmark (JDK11 is slower in general) - [x] DateTimeBenchmark (JDK11 is slightly slower in general except `parsing`) - [x] MakeDateTimeBenchmark (JDK11 is slower except two cases) - [x] MiscBenchmark (JDK11 is slower except ten cases) - [x] OrcV2NestedSchemaPruningBenchmark (JDK11 is slower) - [x] ParquetNestedSchemaPruningBenchmark (JDK11 is slower except six cases) - [x] RangeBenchmark (JDK11 is slower except one case) `FilterPushdownBenchmark/InExpressionBenchmark/WideSchemaBenchmark` will be compared later because it took long timer. ### Why are the changes needed? According to the result, there are some difference between JDK8/JDK11. This will be a baseline for the future improvement and comparison. Also, as a reproducible environment, the following environment is used. - Instance: `r3.xlarge` - OS: `CentOS Linux release 7.5.1804 (Core)` - JDK: - `OpenJDK Runtime Environment (build 1.8.0_222-b10)` - `OpenJDK Runtime Environment 18.9 (build 11.0.4+11-LTS)` ### Does this PR introduce any user-facing change? No. ### How was this patch tested? This is a test-only PR. We need to run benchmark. Closes #26003 from dongjoon-hyun/SPARK-29320. Authored-by: Dongjoon Hyun <dhyun@apple.com> Signed-off-by: Dongjoon Hyun <dhyun@apple.com>
2019-10-03 11:58:25 -04:00
[SPARK-31527][SQL][TESTS][FOLLOWUP] Fix the number of rows in `DateTimeBenchmark` ### What changes were proposed in this pull request? - Changed to the number of rows in benchmark cases from 3 to the actual number `N`. - Regenerated benchmark results in the environment: | Item | Description | | ---- | ----| | Region | us-west-2 (Oregon) | | Instance | r3.xlarge | | AMI | ubuntu/images/hvm-ssd/ubuntu-bionic-18.04-amd64-server-20190722.1 (ami-06f2f779464715dc5) | | Java | OpenJDK 64-Bit Server VM 1.8.0_242 and OpenJDK 64-Bit Server VM 11.0.6+10 | ### Why are the changes needed? The changes are needed to have: - Correct benchmark results - Base line for other perf improvements that can be checked in the same environment. ### Does this PR introduce _any_ user-facing change? No ### How was this patch tested? By running the benchmark and checking its output. Closes #28440 from MaxGekk/SPARK-31527-DateTimeBenchmark-followup. Authored-by: Max Gekk <max.gekk@gmail.com> Signed-off-by: Takeshi Yamamuro <yamamuro@apache.org>
2020-05-03 20:39:50 -04:00
OpenJDK 64-Bit Server VM 11.0.7+10-post-Ubuntu-2ubuntu218.04 on Linux 4.15.0-1063-aws
Intel(R) Xeon(R) CPU E5-2670 v2 @ 2.50GHz
[SPARK-29320][TESTS] Compare `sql/core` module in JDK8/11 (Part 1) ### What changes were proposed in this pull request? This PR regenerates the `sql/core` benchmarks in JDK8/11 to compare the result. In general, we compare the ratio instead of the time. However, in this PR, the average time is compared. This PR should be considered as a rough comparison. **A. EXPECTED CASES(JDK11 is faster in general)** - [x] BloomFilterBenchmark (JDK11 is faster except one case) - [x] BuiltInDataSourceWriteBenchmark (JDK11 is faster at CSV/ORC) - [x] CSVBenchmark (JDK11 is faster except five cases) - [x] ColumnarBatchBenchmark (JDK11 is faster at `boolean`/`string` and some cases in `int`/`array`) - [x] DatasetBenchmark (JDK11 is faster with `string`, but is slower for `long` type) - [x] ExternalAppendOnlyUnsafeRowArrayBenchmark (JDK11 is faster except two cases) - [x] ExtractBenchmark (JDK11 is faster except HOUR/MINUTE/SECOND/MILLISECONDS/MICROSECONDS) - [x] HashedRelationMetricsBenchmark (JDK11 is faster) - [x] JSONBenchmark (JDK11 is much faster except eight cases) - [x] JoinBenchmark (JDK11 is faster except five cases) - [x] OrcNestedSchemaPruningBenchmark (JDK11 is faster in nine cases) - [x] PrimitiveArrayBenchmark (JDK11 is faster) - [x] SortBenchmark (JDK11 is faster except `Arrays.sort` case) - [x] UDFBenchmark (N/A, values are too small) - [x] UnsafeArrayDataBenchmark (JDK11 is faster except one case) - [x] WideTableBenchmark (JDK11 is faster except two cases) **B. CASES WE NEED TO INVESTIGATE MORE LATER** - [x] AggregateBenchmark (JDK11 is slower in general) - [x] CompressionSchemeBenchmark (JDK11 is slower in general except `string`) - [x] DataSourceReadBenchmark (JDK11 is slower in general) - [x] DateTimeBenchmark (JDK11 is slightly slower in general except `parsing`) - [x] MakeDateTimeBenchmark (JDK11 is slower except two cases) - [x] MiscBenchmark (JDK11 is slower except ten cases) - [x] OrcV2NestedSchemaPruningBenchmark (JDK11 is slower) - [x] ParquetNestedSchemaPruningBenchmark (JDK11 is slower except six cases) - [x] RangeBenchmark (JDK11 is slower except one case) `FilterPushdownBenchmark/InExpressionBenchmark/WideSchemaBenchmark` will be compared later because it took long timer. ### Why are the changes needed? According to the result, there are some difference between JDK8/JDK11. This will be a baseline for the future improvement and comparison. Also, as a reproducible environment, the following environment is used. - Instance: `r3.xlarge` - OS: `CentOS Linux release 7.5.1804 (Core)` - JDK: - `OpenJDK Runtime Environment (build 1.8.0_222-b10)` - `OpenJDK Runtime Environment 18.9 (build 11.0.4+11-LTS)` ### Does this PR introduce any user-facing change? No. ### How was this patch tested? This is a test-only PR. We need to run benchmark. Closes #26003 from dongjoon-hyun/SPARK-29320. Authored-by: Dongjoon Hyun <dhyun@apple.com> Signed-off-by: Dongjoon Hyun <dhyun@apple.com>
2019-10-03 11:58:25 -04:00
to_date: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative
------------------------------------------------------------------------------------------------------------------------
[SPARK-31630][SQL] Fix perf regression by skipping timestamps rebasing after some threshold ### What changes were proposed in this pull request? Skip timestamps rebasing after a global threshold when there is no difference between Julian and Gregorian calendars. This allows to avoid checking hash maps of switch points, and fixes perf regressions in `toJavaTimestamp()` and `fromJavaTimestamp()`. ### Why are the changes needed? The changes fix perf regressions of conversions to/from external type `java.sql.Timestamp`. Before (see the PR's results https://github.com/apache/spark/pull/28440): ``` ================================================================================================ Conversion from/to external types ================================================================================================ OpenJDK 64-Bit Server VM 1.8.0_252-8u252-b09-1~18.04-b09 on Linux 4.15.0-1063-aws Intel(R) Xeon(R) CPU E5-2670 v2 2.50GHz To/from Java's date-time: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative ------------------------------------------------------------------------------------------------------------------------ From java.sql.Timestamp 376 388 10 13.3 75.2 1.1X Collect java.sql.Timestamp 1878 1937 64 2.7 375.6 0.2X ``` After: ``` ================================================================================================ Conversion from/to external types ================================================================================================ OpenJDK 64-Bit Server VM 1.8.0_252-8u252-b09-1~18.04-b09 on Linux 4.15.0-1063-aws Intel(R) Xeon(R) CPU E5-2670 v2 2.50GHz To/from Java's date-time: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative ------------------------------------------------------------------------------------------------------------------------ From java.sql.Timestamp 249 264 24 20.1 49.8 1.7X Collect java.sql.Timestamp 1503 1523 24 3.3 300.5 0.3X ``` Perf improvements in average of: 1. From java.sql.Timestamp is ~ 34% 2. To java.sql.Timestamps is ~16% ### Does this PR introduce _any_ user-facing change? No ### How was this patch tested? By existing test suites `DateTimeUtilsSuite` and `RebaseDateTimeSuite`. Closes #28441 from MaxGekk/opt-rebase-common-threshold. Authored-by: Max Gekk <max.gekk@gmail.com> Signed-off-by: Wenchen Fan <wenchen@databricks.com>
2020-05-05 10:11:53 -04:00
to_date wholestage off 3413 3440 38 0.3 3413.0 1.0X
to_date wholestage on 3392 3402 12 0.3 3392.3 1.0X
[SPARK-29320][TESTS] Compare `sql/core` module in JDK8/11 (Part 1) ### What changes were proposed in this pull request? This PR regenerates the `sql/core` benchmarks in JDK8/11 to compare the result. In general, we compare the ratio instead of the time. However, in this PR, the average time is compared. This PR should be considered as a rough comparison. **A. EXPECTED CASES(JDK11 is faster in general)** - [x] BloomFilterBenchmark (JDK11 is faster except one case) - [x] BuiltInDataSourceWriteBenchmark (JDK11 is faster at CSV/ORC) - [x] CSVBenchmark (JDK11 is faster except five cases) - [x] ColumnarBatchBenchmark (JDK11 is faster at `boolean`/`string` and some cases in `int`/`array`) - [x] DatasetBenchmark (JDK11 is faster with `string`, but is slower for `long` type) - [x] ExternalAppendOnlyUnsafeRowArrayBenchmark (JDK11 is faster except two cases) - [x] ExtractBenchmark (JDK11 is faster except HOUR/MINUTE/SECOND/MILLISECONDS/MICROSECONDS) - [x] HashedRelationMetricsBenchmark (JDK11 is faster) - [x] JSONBenchmark (JDK11 is much faster except eight cases) - [x] JoinBenchmark (JDK11 is faster except five cases) - [x] OrcNestedSchemaPruningBenchmark (JDK11 is faster in nine cases) - [x] PrimitiveArrayBenchmark (JDK11 is faster) - [x] SortBenchmark (JDK11 is faster except `Arrays.sort` case) - [x] UDFBenchmark (N/A, values are too small) - [x] UnsafeArrayDataBenchmark (JDK11 is faster except one case) - [x] WideTableBenchmark (JDK11 is faster except two cases) **B. CASES WE NEED TO INVESTIGATE MORE LATER** - [x] AggregateBenchmark (JDK11 is slower in general) - [x] CompressionSchemeBenchmark (JDK11 is slower in general except `string`) - [x] DataSourceReadBenchmark (JDK11 is slower in general) - [x] DateTimeBenchmark (JDK11 is slightly slower in general except `parsing`) - [x] MakeDateTimeBenchmark (JDK11 is slower except two cases) - [x] MiscBenchmark (JDK11 is slower except ten cases) - [x] OrcV2NestedSchemaPruningBenchmark (JDK11 is slower) - [x] ParquetNestedSchemaPruningBenchmark (JDK11 is slower except six cases) - [x] RangeBenchmark (JDK11 is slower except one case) `FilterPushdownBenchmark/InExpressionBenchmark/WideSchemaBenchmark` will be compared later because it took long timer. ### Why are the changes needed? According to the result, there are some difference between JDK8/JDK11. This will be a baseline for the future improvement and comparison. Also, as a reproducible environment, the following environment is used. - Instance: `r3.xlarge` - OS: `CentOS Linux release 7.5.1804 (Core)` - JDK: - `OpenJDK Runtime Environment (build 1.8.0_222-b10)` - `OpenJDK Runtime Environment 18.9 (build 11.0.4+11-LTS)` ### Does this PR introduce any user-facing change? No. ### How was this patch tested? This is a test-only PR. We need to run benchmark. Closes #26003 from dongjoon-hyun/SPARK-29320. Authored-by: Dongjoon Hyun <dhyun@apple.com> Signed-off-by: Dongjoon Hyun <dhyun@apple.com>
2019-10-03 11:58:25 -04:00
================================================================================================
Conversion from/to external types
================================================================================================
[SPARK-31527][SQL][TESTS][FOLLOWUP] Fix the number of rows in `DateTimeBenchmark` ### What changes were proposed in this pull request? - Changed to the number of rows in benchmark cases from 3 to the actual number `N`. - Regenerated benchmark results in the environment: | Item | Description | | ---- | ----| | Region | us-west-2 (Oregon) | | Instance | r3.xlarge | | AMI | ubuntu/images/hvm-ssd/ubuntu-bionic-18.04-amd64-server-20190722.1 (ami-06f2f779464715dc5) | | Java | OpenJDK 64-Bit Server VM 1.8.0_242 and OpenJDK 64-Bit Server VM 11.0.6+10 | ### Why are the changes needed? The changes are needed to have: - Correct benchmark results - Base line for other perf improvements that can be checked in the same environment. ### Does this PR introduce _any_ user-facing change? No ### How was this patch tested? By running the benchmark and checking its output. Closes #28440 from MaxGekk/SPARK-31527-DateTimeBenchmark-followup. Authored-by: Max Gekk <max.gekk@gmail.com> Signed-off-by: Takeshi Yamamuro <yamamuro@apache.org>
2020-05-03 20:39:50 -04:00
OpenJDK 64-Bit Server VM 11.0.7+10-post-Ubuntu-2ubuntu218.04 on Linux 4.15.0-1063-aws
Intel(R) Xeon(R) CPU E5-2670 v2 @ 2.50GHz
[SPARK-31439][SQL] Fix perf regression of fromJavaDate ### What changes were proposed in this pull request? In the PR, I propose to re-use optimized implementation of days rebase function `rebaseJulianToGregorianDays()` introduced by the PR #28067 in conversion of `java.sql.Date` values to Catalyst's `DATE` values. The function `fromJavaDate` in `DateTimeUtils` was re-written by taking the implementation from Spark 2.4, and by rebasing the final results via `rebaseJulianToGregorianDays()`. Also I updated `DateTimeBenchmark`, and added a benchmark for conversion from `java.sql.Date`. ### Why are the changes needed? The PR fixes the regression of parallelizing a collection of `java.sql.Date` values, and improves performance of converting external values to Catalyst's `DATE` values: - x4 on the master branch - 30% against Spark 2.4.6-SNAPSHOT Spark 2.4.6-SNAPSHOT: ``` To/from java.sql.Timestamp: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative ------------------------------------------------------------------------------------------------------------------------ From java.sql.Date 614 655 43 8.1 122.8 1.0X ``` Before the changes: ``` To/from java.sql.Timestamp: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative ------------------------------------------------------------------------------------------------------------------------ From java.sql.Date 1154 1206 46 4.3 230.9 1.0X ``` After: ``` To/from java.sql.Timestamp: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative ------------------------------------------------------------------------------------------------------------------------ From java.sql.Date 427 434 7 11.7 85.3 1.0X ``` ### Does this PR introduce any user-facing change? No ### How was this patch tested? - By existing tests suites, in particular, `DateTimeUtilsSuite`, `RebaseDateTimeSuite`, `DateFunctionsSuite`, `DateExpressionsSuite`. - Re-run `DateTimeBenchmark` in the environment: | Item | Description | | ---- | ----| | Region | us-west-2 (Oregon) | | Instance | r3.xlarge | | AMI | ubuntu/images/hvm-ssd/ubuntu-bionic-18.04-amd64-server-20190722.1 (ami-06f2f779464715dc5) | | Java | OpenJDK 64-Bit Server VM 1.8.0_242 and OpenJDK 64-Bit Server VM 11.0.6+10 | Closes #28205 from MaxGekk/optimize-fromJavaDate. Authored-by: Max Gekk <max.gekk@gmail.com> Signed-off-by: Wenchen Fan <wenchen@databricks.com>
2020-04-14 10:44:00 -04:00
To/from Java's date-time: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative
[SPARK-29320][TESTS] Compare `sql/core` module in JDK8/11 (Part 1) ### What changes were proposed in this pull request? This PR regenerates the `sql/core` benchmarks in JDK8/11 to compare the result. In general, we compare the ratio instead of the time. However, in this PR, the average time is compared. This PR should be considered as a rough comparison. **A. EXPECTED CASES(JDK11 is faster in general)** - [x] BloomFilterBenchmark (JDK11 is faster except one case) - [x] BuiltInDataSourceWriteBenchmark (JDK11 is faster at CSV/ORC) - [x] CSVBenchmark (JDK11 is faster except five cases) - [x] ColumnarBatchBenchmark (JDK11 is faster at `boolean`/`string` and some cases in `int`/`array`) - [x] DatasetBenchmark (JDK11 is faster with `string`, but is slower for `long` type) - [x] ExternalAppendOnlyUnsafeRowArrayBenchmark (JDK11 is faster except two cases) - [x] ExtractBenchmark (JDK11 is faster except HOUR/MINUTE/SECOND/MILLISECONDS/MICROSECONDS) - [x] HashedRelationMetricsBenchmark (JDK11 is faster) - [x] JSONBenchmark (JDK11 is much faster except eight cases) - [x] JoinBenchmark (JDK11 is faster except five cases) - [x] OrcNestedSchemaPruningBenchmark (JDK11 is faster in nine cases) - [x] PrimitiveArrayBenchmark (JDK11 is faster) - [x] SortBenchmark (JDK11 is faster except `Arrays.sort` case) - [x] UDFBenchmark (N/A, values are too small) - [x] UnsafeArrayDataBenchmark (JDK11 is faster except one case) - [x] WideTableBenchmark (JDK11 is faster except two cases) **B. CASES WE NEED TO INVESTIGATE MORE LATER** - [x] AggregateBenchmark (JDK11 is slower in general) - [x] CompressionSchemeBenchmark (JDK11 is slower in general except `string`) - [x] DataSourceReadBenchmark (JDK11 is slower in general) - [x] DateTimeBenchmark (JDK11 is slightly slower in general except `parsing`) - [x] MakeDateTimeBenchmark (JDK11 is slower except two cases) - [x] MiscBenchmark (JDK11 is slower except ten cases) - [x] OrcV2NestedSchemaPruningBenchmark (JDK11 is slower) - [x] ParquetNestedSchemaPruningBenchmark (JDK11 is slower except six cases) - [x] RangeBenchmark (JDK11 is slower except one case) `FilterPushdownBenchmark/InExpressionBenchmark/WideSchemaBenchmark` will be compared later because it took long timer. ### Why are the changes needed? According to the result, there are some difference between JDK8/JDK11. This will be a baseline for the future improvement and comparison. Also, as a reproducible environment, the following environment is used. - Instance: `r3.xlarge` - OS: `CentOS Linux release 7.5.1804 (Core)` - JDK: - `OpenJDK Runtime Environment (build 1.8.0_222-b10)` - `OpenJDK Runtime Environment 18.9 (build 11.0.4+11-LTS)` ### Does this PR introduce any user-facing change? No. ### How was this patch tested? This is a test-only PR. We need to run benchmark. Closes #26003 from dongjoon-hyun/SPARK-29320. Authored-by: Dongjoon Hyun <dhyun@apple.com> Signed-off-by: Dongjoon Hyun <dhyun@apple.com>
2019-10-03 11:58:25 -04:00
------------------------------------------------------------------------------------------------------------------------
[SPARK-31630][SQL] Fix perf regression by skipping timestamps rebasing after some threshold ### What changes were proposed in this pull request? Skip timestamps rebasing after a global threshold when there is no difference between Julian and Gregorian calendars. This allows to avoid checking hash maps of switch points, and fixes perf regressions in `toJavaTimestamp()` and `fromJavaTimestamp()`. ### Why are the changes needed? The changes fix perf regressions of conversions to/from external type `java.sql.Timestamp`. Before (see the PR's results https://github.com/apache/spark/pull/28440): ``` ================================================================================================ Conversion from/to external types ================================================================================================ OpenJDK 64-Bit Server VM 1.8.0_252-8u252-b09-1~18.04-b09 on Linux 4.15.0-1063-aws Intel(R) Xeon(R) CPU E5-2670 v2 2.50GHz To/from Java's date-time: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative ------------------------------------------------------------------------------------------------------------------------ From java.sql.Timestamp 376 388 10 13.3 75.2 1.1X Collect java.sql.Timestamp 1878 1937 64 2.7 375.6 0.2X ``` After: ``` ================================================================================================ Conversion from/to external types ================================================================================================ OpenJDK 64-Bit Server VM 1.8.0_252-8u252-b09-1~18.04-b09 on Linux 4.15.0-1063-aws Intel(R) Xeon(R) CPU E5-2670 v2 2.50GHz To/from Java's date-time: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative ------------------------------------------------------------------------------------------------------------------------ From java.sql.Timestamp 249 264 24 20.1 49.8 1.7X Collect java.sql.Timestamp 1503 1523 24 3.3 300.5 0.3X ``` Perf improvements in average of: 1. From java.sql.Timestamp is ~ 34% 2. To java.sql.Timestamps is ~16% ### Does this PR introduce _any_ user-facing change? No ### How was this patch tested? By existing test suites `DateTimeUtilsSuite` and `RebaseDateTimeSuite`. Closes #28441 from MaxGekk/opt-rebase-common-threshold. Authored-by: Max Gekk <max.gekk@gmail.com> Signed-off-by: Wenchen Fan <wenchen@databricks.com>
2020-05-05 10:11:53 -04:00
From java.sql.Date 410 415 7 12.2 82.0 1.0X
From java.time.LocalDate 332 333 1 15.1 66.4 1.2X
Collect java.sql.Date 1891 2542 829 2.6 378.1 0.2X
Collect java.time.LocalDate 1630 2138 441 3.1 326.0 0.3X
From java.sql.Timestamp 254 259 6 19.7 50.9 1.6X
From java.time.Instant 302 306 4 16.6 60.3 1.4X
Collect longs 1134 1265 117 4.4 226.8 0.4X
Collect java.sql.Timestamp 1441 1458 16 3.5 288.1 0.3X
Collect java.time.Instant 1680 1928 253 3.0 336.0 0.2X
[SPARK-29320][TESTS] Compare `sql/core` module in JDK8/11 (Part 1) ### What changes were proposed in this pull request? This PR regenerates the `sql/core` benchmarks in JDK8/11 to compare the result. In general, we compare the ratio instead of the time. However, in this PR, the average time is compared. This PR should be considered as a rough comparison. **A. EXPECTED CASES(JDK11 is faster in general)** - [x] BloomFilterBenchmark (JDK11 is faster except one case) - [x] BuiltInDataSourceWriteBenchmark (JDK11 is faster at CSV/ORC) - [x] CSVBenchmark (JDK11 is faster except five cases) - [x] ColumnarBatchBenchmark (JDK11 is faster at `boolean`/`string` and some cases in `int`/`array`) - [x] DatasetBenchmark (JDK11 is faster with `string`, but is slower for `long` type) - [x] ExternalAppendOnlyUnsafeRowArrayBenchmark (JDK11 is faster except two cases) - [x] ExtractBenchmark (JDK11 is faster except HOUR/MINUTE/SECOND/MILLISECONDS/MICROSECONDS) - [x] HashedRelationMetricsBenchmark (JDK11 is faster) - [x] JSONBenchmark (JDK11 is much faster except eight cases) - [x] JoinBenchmark (JDK11 is faster except five cases) - [x] OrcNestedSchemaPruningBenchmark (JDK11 is faster in nine cases) - [x] PrimitiveArrayBenchmark (JDK11 is faster) - [x] SortBenchmark (JDK11 is faster except `Arrays.sort` case) - [x] UDFBenchmark (N/A, values are too small) - [x] UnsafeArrayDataBenchmark (JDK11 is faster except one case) - [x] WideTableBenchmark (JDK11 is faster except two cases) **B. CASES WE NEED TO INVESTIGATE MORE LATER** - [x] AggregateBenchmark (JDK11 is slower in general) - [x] CompressionSchemeBenchmark (JDK11 is slower in general except `string`) - [x] DataSourceReadBenchmark (JDK11 is slower in general) - [x] DateTimeBenchmark (JDK11 is slightly slower in general except `parsing`) - [x] MakeDateTimeBenchmark (JDK11 is slower except two cases) - [x] MiscBenchmark (JDK11 is slower except ten cases) - [x] OrcV2NestedSchemaPruningBenchmark (JDK11 is slower) - [x] ParquetNestedSchemaPruningBenchmark (JDK11 is slower except six cases) - [x] RangeBenchmark (JDK11 is slower except one case) `FilterPushdownBenchmark/InExpressionBenchmark/WideSchemaBenchmark` will be compared later because it took long timer. ### Why are the changes needed? According to the result, there are some difference between JDK8/JDK11. This will be a baseline for the future improvement and comparison. Also, as a reproducible environment, the following environment is used. - Instance: `r3.xlarge` - OS: `CentOS Linux release 7.5.1804 (Core)` - JDK: - `OpenJDK Runtime Environment (build 1.8.0_222-b10)` - `OpenJDK Runtime Environment 18.9 (build 11.0.4+11-LTS)` ### Does this PR introduce any user-facing change? No. ### How was this patch tested? This is a test-only PR. We need to run benchmark. Closes #26003 from dongjoon-hyun/SPARK-29320. Authored-by: Dongjoon Hyun <dhyun@apple.com> Signed-off-by: Dongjoon Hyun <dhyun@apple.com>
2019-10-03 11:58:25 -04:00
Reference in a new issue Copy permalink