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

253 lines
23 KiB
Plaintext
Raw Normal View History

[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
================================================================================================
SQL Single Numeric Column Scan
================================================================================================
[SPARK-30409][SPARK-29173][SQL][TESTS] Use `NoOp` datasource in SQL benchmarks ### What changes were proposed in this pull request? In the PR, I propose to replace `.collect()`, `.count()` and `.foreach(_ => ())` in SQL benchmarks and use the `NoOp` datasource. I added an implicit class to `SqlBasedBenchmark` with the `.noop()` method. It can be used in benchmark like: `ds.noop()`. The last one is unfolded to `ds.write.format("noop").mode(Overwrite).save()`. ### Why are the changes needed? To avoid additional overhead that `collect()` (and other actions) has. For example, `.collect()` has to convert values according to external types and pull data to the driver. This can hide actual performance regressions or improvements of benchmarked operations. ### Does this PR introduce any user-facing change? No ### How was this patch tested? Re-run all modified benchmarks using Amazon EC2. | Item | Description | | ---- | ----| | Region | us-west-2 (Oregon) | | Instance | r3.xlarge (spot instance) | | AMI | ami-06f2f779464715dc5 (ubuntu/images/hvm-ssd/ubuntu-bionic-18.04-amd64-server-20190722.1) | | Java | OpenJDK8/10 | - Run `TPCDSQueryBenchmark` using instructions from the PR #26049 ``` # `spark-tpcds-datagen` needs this. (JDK8) $ git clone https://github.com/apache/spark.git -b branch-2.4 --depth 1 spark-2.4 $ export SPARK_HOME=$PWD $ ./build/mvn clean package -DskipTests # Generate data. (JDK8) $ git clone gitgithub.com:maropu/spark-tpcds-datagen.git $ cd spark-tpcds-datagen/ $ build/mvn clean package $ mkdir -p /data/tpcds $ ./bin/dsdgen --output-location /data/tpcds/s1 // This need `Spark 2.4` ``` - Other benchmarks ran by the script: ``` #!/usr/bin/env python3 import os from sparktestsupport.shellutils import run_cmd benchmarks = [ ['sql/test', 'org.apache.spark.sql.execution.benchmark.AggregateBenchmark'], ['avro/test', 'org.apache.spark.sql.execution.benchmark.AvroReadBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.BloomFilterBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.DataSourceReadBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.DateTimeBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.ExtractBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.FilterPushdownBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.InExpressionBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.IntervalBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.JoinBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.MakeDateTimeBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.MiscBenchmark'], ['hive/test', 'org.apache.spark.sql.execution.benchmark.ObjectHashAggregateExecBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.OrcNestedSchemaPruningBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.OrcV2NestedSchemaPruningBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.ParquetNestedSchemaPruningBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.RangeBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.UDFBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.WideSchemaBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.WideTableBenchmark'], ['hive/test', 'org.apache.spark.sql.hive.orc.OrcReadBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.datasources.csv.CSVBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.datasources.json.JsonBenchmark'] ] print('Set SPARK_GENERATE_BENCHMARK_FILES=1') os.environ['SPARK_GENERATE_BENCHMARK_FILES'] = '1' for b in benchmarks: print("Run benchmark: %s" % b[1]) run_cmd(['build/sbt', '%s:runMain %s' % (b[0], b[1])]) ``` Closes #27078 from MaxGekk/noop-in-benchmarks. Lead-authored-by: Maxim Gekk <max.gekk@gmail.com> Co-authored-by: Maxim Gekk <maxim.gekk@databricks.com> Co-authored-by: Dongjoon Hyun <dhyun@apple.com> Signed-off-by: Dongjoon Hyun <dhyun@apple.com>
2020-01-12 16:18:19 -05:00
OpenJDK 64-Bit Server VM 11.0.5+10-post-Ubuntu-0ubuntu1.118.04 on Linux 4.15.0-1044-aws
[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
Intel(R) Xeon(R) CPU E5-2670 v2 @ 2.50GHz
SQL Single TINYINT Column Scan: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative
------------------------------------------------------------------------------------------------------------------------
[SPARK-30409][SPARK-29173][SQL][TESTS] Use `NoOp` datasource in SQL benchmarks ### What changes were proposed in this pull request? In the PR, I propose to replace `.collect()`, `.count()` and `.foreach(_ => ())` in SQL benchmarks and use the `NoOp` datasource. I added an implicit class to `SqlBasedBenchmark` with the `.noop()` method. It can be used in benchmark like: `ds.noop()`. The last one is unfolded to `ds.write.format("noop").mode(Overwrite).save()`. ### Why are the changes needed? To avoid additional overhead that `collect()` (and other actions) has. For example, `.collect()` has to convert values according to external types and pull data to the driver. This can hide actual performance regressions or improvements of benchmarked operations. ### Does this PR introduce any user-facing change? No ### How was this patch tested? Re-run all modified benchmarks using Amazon EC2. | Item | Description | | ---- | ----| | Region | us-west-2 (Oregon) | | Instance | r3.xlarge (spot instance) | | AMI | ami-06f2f779464715dc5 (ubuntu/images/hvm-ssd/ubuntu-bionic-18.04-amd64-server-20190722.1) | | Java | OpenJDK8/10 | - Run `TPCDSQueryBenchmark` using instructions from the PR #26049 ``` # `spark-tpcds-datagen` needs this. (JDK8) $ git clone https://github.com/apache/spark.git -b branch-2.4 --depth 1 spark-2.4 $ export SPARK_HOME=$PWD $ ./build/mvn clean package -DskipTests # Generate data. (JDK8) $ git clone gitgithub.com:maropu/spark-tpcds-datagen.git $ cd spark-tpcds-datagen/ $ build/mvn clean package $ mkdir -p /data/tpcds $ ./bin/dsdgen --output-location /data/tpcds/s1 // This need `Spark 2.4` ``` - Other benchmarks ran by the script: ``` #!/usr/bin/env python3 import os from sparktestsupport.shellutils import run_cmd benchmarks = [ ['sql/test', 'org.apache.spark.sql.execution.benchmark.AggregateBenchmark'], ['avro/test', 'org.apache.spark.sql.execution.benchmark.AvroReadBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.BloomFilterBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.DataSourceReadBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.DateTimeBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.ExtractBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.FilterPushdownBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.InExpressionBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.IntervalBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.JoinBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.MakeDateTimeBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.MiscBenchmark'], ['hive/test', 'org.apache.spark.sql.execution.benchmark.ObjectHashAggregateExecBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.OrcNestedSchemaPruningBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.OrcV2NestedSchemaPruningBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.ParquetNestedSchemaPruningBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.RangeBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.UDFBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.WideSchemaBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.WideTableBenchmark'], ['hive/test', 'org.apache.spark.sql.hive.orc.OrcReadBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.datasources.csv.CSVBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.datasources.json.JsonBenchmark'] ] print('Set SPARK_GENERATE_BENCHMARK_FILES=1') os.environ['SPARK_GENERATE_BENCHMARK_FILES'] = '1' for b in benchmarks: print("Run benchmark: %s" % b[1]) run_cmd(['build/sbt', '%s:runMain %s' % (b[0], b[1])]) ``` Closes #27078 from MaxGekk/noop-in-benchmarks. Lead-authored-by: Maxim Gekk <max.gekk@gmail.com> Co-authored-by: Maxim Gekk <maxim.gekk@databricks.com> Co-authored-by: Dongjoon Hyun <dhyun@apple.com> Signed-off-by: Dongjoon Hyun <dhyun@apple.com>
2020-01-12 16:18:19 -05:00
SQL CSV 28338 28589 356 0.6 1801.7 1.0X
SQL Json 9273 9332 83 1.7 589.6 3.1X
SQL Parquet Vectorized 186 217 22 84.3 11.9 152.0X
SQL Parquet MR 1951 1972 29 8.1 124.1 14.5X
SQL ORC Vectorized 256 277 22 61.4 16.3 110.6X
SQL ORC MR 1627 1717 127 9.7 103.4 17.4X
[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-30409][SPARK-29173][SQL][TESTS] Use `NoOp` datasource in SQL benchmarks ### What changes were proposed in this pull request? In the PR, I propose to replace `.collect()`, `.count()` and `.foreach(_ => ())` in SQL benchmarks and use the `NoOp` datasource. I added an implicit class to `SqlBasedBenchmark` with the `.noop()` method. It can be used in benchmark like: `ds.noop()`. The last one is unfolded to `ds.write.format("noop").mode(Overwrite).save()`. ### Why are the changes needed? To avoid additional overhead that `collect()` (and other actions) has. For example, `.collect()` has to convert values according to external types and pull data to the driver. This can hide actual performance regressions or improvements of benchmarked operations. ### Does this PR introduce any user-facing change? No ### How was this patch tested? Re-run all modified benchmarks using Amazon EC2. | Item | Description | | ---- | ----| | Region | us-west-2 (Oregon) | | Instance | r3.xlarge (spot instance) | | AMI | ami-06f2f779464715dc5 (ubuntu/images/hvm-ssd/ubuntu-bionic-18.04-amd64-server-20190722.1) | | Java | OpenJDK8/10 | - Run `TPCDSQueryBenchmark` using instructions from the PR #26049 ``` # `spark-tpcds-datagen` needs this. (JDK8) $ git clone https://github.com/apache/spark.git -b branch-2.4 --depth 1 spark-2.4 $ export SPARK_HOME=$PWD $ ./build/mvn clean package -DskipTests # Generate data. (JDK8) $ git clone gitgithub.com:maropu/spark-tpcds-datagen.git $ cd spark-tpcds-datagen/ $ build/mvn clean package $ mkdir -p /data/tpcds $ ./bin/dsdgen --output-location /data/tpcds/s1 // This need `Spark 2.4` ``` - Other benchmarks ran by the script: ``` #!/usr/bin/env python3 import os from sparktestsupport.shellutils import run_cmd benchmarks = [ ['sql/test', 'org.apache.spark.sql.execution.benchmark.AggregateBenchmark'], ['avro/test', 'org.apache.spark.sql.execution.benchmark.AvroReadBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.BloomFilterBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.DataSourceReadBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.DateTimeBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.ExtractBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.FilterPushdownBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.InExpressionBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.IntervalBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.JoinBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.MakeDateTimeBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.MiscBenchmark'], ['hive/test', 'org.apache.spark.sql.execution.benchmark.ObjectHashAggregateExecBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.OrcNestedSchemaPruningBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.OrcV2NestedSchemaPruningBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.ParquetNestedSchemaPruningBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.RangeBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.UDFBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.WideSchemaBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.WideTableBenchmark'], ['hive/test', 'org.apache.spark.sql.hive.orc.OrcReadBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.datasources.csv.CSVBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.datasources.json.JsonBenchmark'] ] print('Set SPARK_GENERATE_BENCHMARK_FILES=1') os.environ['SPARK_GENERATE_BENCHMARK_FILES'] = '1' for b in benchmarks: print("Run benchmark: %s" % b[1]) run_cmd(['build/sbt', '%s:runMain %s' % (b[0], b[1])]) ``` Closes #27078 from MaxGekk/noop-in-benchmarks. Lead-authored-by: Maxim Gekk <max.gekk@gmail.com> Co-authored-by: Maxim Gekk <maxim.gekk@databricks.com> Co-authored-by: Dongjoon Hyun <dhyun@apple.com> Signed-off-by: Dongjoon Hyun <dhyun@apple.com>
2020-01-12 16:18:19 -05:00
OpenJDK 64-Bit Server VM 11.0.5+10-post-Ubuntu-0ubuntu1.118.04 on Linux 4.15.0-1044-aws
[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
Intel(R) Xeon(R) CPU E5-2670 v2 @ 2.50GHz
Parquet Reader Single TINYINT Column Scan: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative
------------------------------------------------------------------------------------------------------------------------
[SPARK-30409][SPARK-29173][SQL][TESTS] Use `NoOp` datasource in SQL benchmarks ### What changes were proposed in this pull request? In the PR, I propose to replace `.collect()`, `.count()` and `.foreach(_ => ())` in SQL benchmarks and use the `NoOp` datasource. I added an implicit class to `SqlBasedBenchmark` with the `.noop()` method. It can be used in benchmark like: `ds.noop()`. The last one is unfolded to `ds.write.format("noop").mode(Overwrite).save()`. ### Why are the changes needed? To avoid additional overhead that `collect()` (and other actions) has. For example, `.collect()` has to convert values according to external types and pull data to the driver. This can hide actual performance regressions or improvements of benchmarked operations. ### Does this PR introduce any user-facing change? No ### How was this patch tested? Re-run all modified benchmarks using Amazon EC2. | Item | Description | | ---- | ----| | Region | us-west-2 (Oregon) | | Instance | r3.xlarge (spot instance) | | AMI | ami-06f2f779464715dc5 (ubuntu/images/hvm-ssd/ubuntu-bionic-18.04-amd64-server-20190722.1) | | Java | OpenJDK8/10 | - Run `TPCDSQueryBenchmark` using instructions from the PR #26049 ``` # `spark-tpcds-datagen` needs this. (JDK8) $ git clone https://github.com/apache/spark.git -b branch-2.4 --depth 1 spark-2.4 $ export SPARK_HOME=$PWD $ ./build/mvn clean package -DskipTests # Generate data. (JDK8) $ git clone gitgithub.com:maropu/spark-tpcds-datagen.git $ cd spark-tpcds-datagen/ $ build/mvn clean package $ mkdir -p /data/tpcds $ ./bin/dsdgen --output-location /data/tpcds/s1 // This need `Spark 2.4` ``` - Other benchmarks ran by the script: ``` #!/usr/bin/env python3 import os from sparktestsupport.shellutils import run_cmd benchmarks = [ ['sql/test', 'org.apache.spark.sql.execution.benchmark.AggregateBenchmark'], ['avro/test', 'org.apache.spark.sql.execution.benchmark.AvroReadBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.BloomFilterBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.DataSourceReadBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.DateTimeBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.ExtractBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.FilterPushdownBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.InExpressionBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.IntervalBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.JoinBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.MakeDateTimeBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.MiscBenchmark'], ['hive/test', 'org.apache.spark.sql.execution.benchmark.ObjectHashAggregateExecBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.OrcNestedSchemaPruningBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.OrcV2NestedSchemaPruningBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.ParquetNestedSchemaPruningBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.RangeBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.UDFBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.WideSchemaBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.WideTableBenchmark'], ['hive/test', 'org.apache.spark.sql.hive.orc.OrcReadBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.datasources.csv.CSVBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.datasources.json.JsonBenchmark'] ] print('Set SPARK_GENERATE_BENCHMARK_FILES=1') os.environ['SPARK_GENERATE_BENCHMARK_FILES'] = '1' for b in benchmarks: print("Run benchmark: %s" % b[1]) run_cmd(['build/sbt', '%s:runMain %s' % (b[0], b[1])]) ``` Closes #27078 from MaxGekk/noop-in-benchmarks. Lead-authored-by: Maxim Gekk <max.gekk@gmail.com> Co-authored-by: Maxim Gekk <maxim.gekk@databricks.com> Co-authored-by: Dongjoon Hyun <dhyun@apple.com> Signed-off-by: Dongjoon Hyun <dhyun@apple.com>
2020-01-12 16:18:19 -05:00
ParquetReader Vectorized 208 223 13 75.8 13.2 1.0X
ParquetReader Vectorized -> Row 96 97 1 164.1 6.1 2.2X
[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-30409][SPARK-29173][SQL][TESTS] Use `NoOp` datasource in SQL benchmarks ### What changes were proposed in this pull request? In the PR, I propose to replace `.collect()`, `.count()` and `.foreach(_ => ())` in SQL benchmarks and use the `NoOp` datasource. I added an implicit class to `SqlBasedBenchmark` with the `.noop()` method. It can be used in benchmark like: `ds.noop()`. The last one is unfolded to `ds.write.format("noop").mode(Overwrite).save()`. ### Why are the changes needed? To avoid additional overhead that `collect()` (and other actions) has. For example, `.collect()` has to convert values according to external types and pull data to the driver. This can hide actual performance regressions or improvements of benchmarked operations. ### Does this PR introduce any user-facing change? No ### How was this patch tested? Re-run all modified benchmarks using Amazon EC2. | Item | Description | | ---- | ----| | Region | us-west-2 (Oregon) | | Instance | r3.xlarge (spot instance) | | AMI | ami-06f2f779464715dc5 (ubuntu/images/hvm-ssd/ubuntu-bionic-18.04-amd64-server-20190722.1) | | Java | OpenJDK8/10 | - Run `TPCDSQueryBenchmark` using instructions from the PR #26049 ``` # `spark-tpcds-datagen` needs this. (JDK8) $ git clone https://github.com/apache/spark.git -b branch-2.4 --depth 1 spark-2.4 $ export SPARK_HOME=$PWD $ ./build/mvn clean package -DskipTests # Generate data. (JDK8) $ git clone gitgithub.com:maropu/spark-tpcds-datagen.git $ cd spark-tpcds-datagen/ $ build/mvn clean package $ mkdir -p /data/tpcds $ ./bin/dsdgen --output-location /data/tpcds/s1 // This need `Spark 2.4` ``` - Other benchmarks ran by the script: ``` #!/usr/bin/env python3 import os from sparktestsupport.shellutils import run_cmd benchmarks = [ ['sql/test', 'org.apache.spark.sql.execution.benchmark.AggregateBenchmark'], ['avro/test', 'org.apache.spark.sql.execution.benchmark.AvroReadBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.BloomFilterBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.DataSourceReadBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.DateTimeBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.ExtractBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.FilterPushdownBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.InExpressionBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.IntervalBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.JoinBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.MakeDateTimeBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.MiscBenchmark'], ['hive/test', 'org.apache.spark.sql.execution.benchmark.ObjectHashAggregateExecBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.OrcNestedSchemaPruningBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.OrcV2NestedSchemaPruningBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.ParquetNestedSchemaPruningBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.RangeBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.UDFBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.WideSchemaBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.WideTableBenchmark'], ['hive/test', 'org.apache.spark.sql.hive.orc.OrcReadBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.datasources.csv.CSVBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.datasources.json.JsonBenchmark'] ] print('Set SPARK_GENERATE_BENCHMARK_FILES=1') os.environ['SPARK_GENERATE_BENCHMARK_FILES'] = '1' for b in benchmarks: print("Run benchmark: %s" % b[1]) run_cmd(['build/sbt', '%s:runMain %s' % (b[0], b[1])]) ``` Closes #27078 from MaxGekk/noop-in-benchmarks. Lead-authored-by: Maxim Gekk <max.gekk@gmail.com> Co-authored-by: Maxim Gekk <maxim.gekk@databricks.com> Co-authored-by: Dongjoon Hyun <dhyun@apple.com> Signed-off-by: Dongjoon Hyun <dhyun@apple.com>
2020-01-12 16:18:19 -05:00
OpenJDK 64-Bit Server VM 11.0.5+10-post-Ubuntu-0ubuntu1.118.04 on Linux 4.15.0-1044-aws
[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
Intel(R) Xeon(R) CPU E5-2670 v2 @ 2.50GHz
SQL Single SMALLINT Column Scan: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative
------------------------------------------------------------------------------------------------------------------------
[SPARK-30409][SPARK-29173][SQL][TESTS] Use `NoOp` datasource in SQL benchmarks ### What changes were proposed in this pull request? In the PR, I propose to replace `.collect()`, `.count()` and `.foreach(_ => ())` in SQL benchmarks and use the `NoOp` datasource. I added an implicit class to `SqlBasedBenchmark` with the `.noop()` method. It can be used in benchmark like: `ds.noop()`. The last one is unfolded to `ds.write.format("noop").mode(Overwrite).save()`. ### Why are the changes needed? To avoid additional overhead that `collect()` (and other actions) has. For example, `.collect()` has to convert values according to external types and pull data to the driver. This can hide actual performance regressions or improvements of benchmarked operations. ### Does this PR introduce any user-facing change? No ### How was this patch tested? Re-run all modified benchmarks using Amazon EC2. | Item | Description | | ---- | ----| | Region | us-west-2 (Oregon) | | Instance | r3.xlarge (spot instance) | | AMI | ami-06f2f779464715dc5 (ubuntu/images/hvm-ssd/ubuntu-bionic-18.04-amd64-server-20190722.1) | | Java | OpenJDK8/10 | - Run `TPCDSQueryBenchmark` using instructions from the PR #26049 ``` # `spark-tpcds-datagen` needs this. (JDK8) $ git clone https://github.com/apache/spark.git -b branch-2.4 --depth 1 spark-2.4 $ export SPARK_HOME=$PWD $ ./build/mvn clean package -DskipTests # Generate data. (JDK8) $ git clone gitgithub.com:maropu/spark-tpcds-datagen.git $ cd spark-tpcds-datagen/ $ build/mvn clean package $ mkdir -p /data/tpcds $ ./bin/dsdgen --output-location /data/tpcds/s1 // This need `Spark 2.4` ``` - Other benchmarks ran by the script: ``` #!/usr/bin/env python3 import os from sparktestsupport.shellutils import run_cmd benchmarks = [ ['sql/test', 'org.apache.spark.sql.execution.benchmark.AggregateBenchmark'], ['avro/test', 'org.apache.spark.sql.execution.benchmark.AvroReadBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.BloomFilterBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.DataSourceReadBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.DateTimeBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.ExtractBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.FilterPushdownBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.InExpressionBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.IntervalBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.JoinBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.MakeDateTimeBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.MiscBenchmark'], ['hive/test', 'org.apache.spark.sql.execution.benchmark.ObjectHashAggregateExecBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.OrcNestedSchemaPruningBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.OrcV2NestedSchemaPruningBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.ParquetNestedSchemaPruningBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.RangeBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.UDFBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.WideSchemaBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.WideTableBenchmark'], ['hive/test', 'org.apache.spark.sql.hive.orc.OrcReadBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.datasources.csv.CSVBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.datasources.json.JsonBenchmark'] ] print('Set SPARK_GENERATE_BENCHMARK_FILES=1') os.environ['SPARK_GENERATE_BENCHMARK_FILES'] = '1' for b in benchmarks: print("Run benchmark: %s" % b[1]) run_cmd(['build/sbt', '%s:runMain %s' % (b[0], b[1])]) ``` Closes #27078 from MaxGekk/noop-in-benchmarks. Lead-authored-by: Maxim Gekk <max.gekk@gmail.com> Co-authored-by: Maxim Gekk <maxim.gekk@databricks.com> Co-authored-by: Dongjoon Hyun <dhyun@apple.com> Signed-off-by: Dongjoon Hyun <dhyun@apple.com>
2020-01-12 16:18:19 -05:00
SQL CSV 28493 28516 33 0.6 1811.5 1.0X
SQL Json 10257 10291 47 1.5 652.1 2.8X
SQL Parquet Vectorized 215 233 14 73.2 13.7 132.5X
SQL Parquet MR 2384 2388 7 6.6 151.5 12.0X
SQL ORC Vectorized 298 307 7 52.8 18.9 95.6X
SQL ORC MR 1798 1814 22 8.7 114.3 15.8X
[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-30409][SPARK-29173][SQL][TESTS] Use `NoOp` datasource in SQL benchmarks ### What changes were proposed in this pull request? In the PR, I propose to replace `.collect()`, `.count()` and `.foreach(_ => ())` in SQL benchmarks and use the `NoOp` datasource. I added an implicit class to `SqlBasedBenchmark` with the `.noop()` method. It can be used in benchmark like: `ds.noop()`. The last one is unfolded to `ds.write.format("noop").mode(Overwrite).save()`. ### Why are the changes needed? To avoid additional overhead that `collect()` (and other actions) has. For example, `.collect()` has to convert values according to external types and pull data to the driver. This can hide actual performance regressions or improvements of benchmarked operations. ### Does this PR introduce any user-facing change? No ### How was this patch tested? Re-run all modified benchmarks using Amazon EC2. | Item | Description | | ---- | ----| | Region | us-west-2 (Oregon) | | Instance | r3.xlarge (spot instance) | | AMI | ami-06f2f779464715dc5 (ubuntu/images/hvm-ssd/ubuntu-bionic-18.04-amd64-server-20190722.1) | | Java | OpenJDK8/10 | - Run `TPCDSQueryBenchmark` using instructions from the PR #26049 ``` # `spark-tpcds-datagen` needs this. (JDK8) $ git clone https://github.com/apache/spark.git -b branch-2.4 --depth 1 spark-2.4 $ export SPARK_HOME=$PWD $ ./build/mvn clean package -DskipTests # Generate data. (JDK8) $ git clone gitgithub.com:maropu/spark-tpcds-datagen.git $ cd spark-tpcds-datagen/ $ build/mvn clean package $ mkdir -p /data/tpcds $ ./bin/dsdgen --output-location /data/tpcds/s1 // This need `Spark 2.4` ``` - Other benchmarks ran by the script: ``` #!/usr/bin/env python3 import os from sparktestsupport.shellutils import run_cmd benchmarks = [ ['sql/test', 'org.apache.spark.sql.execution.benchmark.AggregateBenchmark'], ['avro/test', 'org.apache.spark.sql.execution.benchmark.AvroReadBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.BloomFilterBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.DataSourceReadBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.DateTimeBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.ExtractBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.FilterPushdownBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.InExpressionBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.IntervalBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.JoinBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.MakeDateTimeBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.MiscBenchmark'], ['hive/test', 'org.apache.spark.sql.execution.benchmark.ObjectHashAggregateExecBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.OrcNestedSchemaPruningBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.OrcV2NestedSchemaPruningBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.ParquetNestedSchemaPruningBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.RangeBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.UDFBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.WideSchemaBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.WideTableBenchmark'], ['hive/test', 'org.apache.spark.sql.hive.orc.OrcReadBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.datasources.csv.CSVBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.datasources.json.JsonBenchmark'] ] print('Set SPARK_GENERATE_BENCHMARK_FILES=1') os.environ['SPARK_GENERATE_BENCHMARK_FILES'] = '1' for b in benchmarks: print("Run benchmark: %s" % b[1]) run_cmd(['build/sbt', '%s:runMain %s' % (b[0], b[1])]) ``` Closes #27078 from MaxGekk/noop-in-benchmarks. Lead-authored-by: Maxim Gekk <max.gekk@gmail.com> Co-authored-by: Maxim Gekk <maxim.gekk@databricks.com> Co-authored-by: Dongjoon Hyun <dhyun@apple.com> Signed-off-by: Dongjoon Hyun <dhyun@apple.com>
2020-01-12 16:18:19 -05:00
OpenJDK 64-Bit Server VM 11.0.5+10-post-Ubuntu-0ubuntu1.118.04 on Linux 4.15.0-1044-aws
[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
Intel(R) Xeon(R) CPU E5-2670 v2 @ 2.50GHz
Parquet Reader Single SMALLINT Column Scan: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative
------------------------------------------------------------------------------------------------------------------------
[SPARK-30409][SPARK-29173][SQL][TESTS] Use `NoOp` datasource in SQL benchmarks ### What changes were proposed in this pull request? In the PR, I propose to replace `.collect()`, `.count()` and `.foreach(_ => ())` in SQL benchmarks and use the `NoOp` datasource. I added an implicit class to `SqlBasedBenchmark` with the `.noop()` method. It can be used in benchmark like: `ds.noop()`. The last one is unfolded to `ds.write.format("noop").mode(Overwrite).save()`. ### Why are the changes needed? To avoid additional overhead that `collect()` (and other actions) has. For example, `.collect()` has to convert values according to external types and pull data to the driver. This can hide actual performance regressions or improvements of benchmarked operations. ### Does this PR introduce any user-facing change? No ### How was this patch tested? Re-run all modified benchmarks using Amazon EC2. | Item | Description | | ---- | ----| | Region | us-west-2 (Oregon) | | Instance | r3.xlarge (spot instance) | | AMI | ami-06f2f779464715dc5 (ubuntu/images/hvm-ssd/ubuntu-bionic-18.04-amd64-server-20190722.1) | | Java | OpenJDK8/10 | - Run `TPCDSQueryBenchmark` using instructions from the PR #26049 ``` # `spark-tpcds-datagen` needs this. (JDK8) $ git clone https://github.com/apache/spark.git -b branch-2.4 --depth 1 spark-2.4 $ export SPARK_HOME=$PWD $ ./build/mvn clean package -DskipTests # Generate data. (JDK8) $ git clone gitgithub.com:maropu/spark-tpcds-datagen.git $ cd spark-tpcds-datagen/ $ build/mvn clean package $ mkdir -p /data/tpcds $ ./bin/dsdgen --output-location /data/tpcds/s1 // This need `Spark 2.4` ``` - Other benchmarks ran by the script: ``` #!/usr/bin/env python3 import os from sparktestsupport.shellutils import run_cmd benchmarks = [ ['sql/test', 'org.apache.spark.sql.execution.benchmark.AggregateBenchmark'], ['avro/test', 'org.apache.spark.sql.execution.benchmark.AvroReadBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.BloomFilterBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.DataSourceReadBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.DateTimeBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.ExtractBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.FilterPushdownBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.InExpressionBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.IntervalBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.JoinBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.MakeDateTimeBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.MiscBenchmark'], ['hive/test', 'org.apache.spark.sql.execution.benchmark.ObjectHashAggregateExecBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.OrcNestedSchemaPruningBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.OrcV2NestedSchemaPruningBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.ParquetNestedSchemaPruningBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.RangeBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.UDFBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.WideSchemaBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.WideTableBenchmark'], ['hive/test', 'org.apache.spark.sql.hive.orc.OrcReadBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.datasources.csv.CSVBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.datasources.json.JsonBenchmark'] ] print('Set SPARK_GENERATE_BENCHMARK_FILES=1') os.environ['SPARK_GENERATE_BENCHMARK_FILES'] = '1' for b in benchmarks: print("Run benchmark: %s" % b[1]) run_cmd(['build/sbt', '%s:runMain %s' % (b[0], b[1])]) ``` Closes #27078 from MaxGekk/noop-in-benchmarks. Lead-authored-by: Maxim Gekk <max.gekk@gmail.com> Co-authored-by: Maxim Gekk <maxim.gekk@databricks.com> Co-authored-by: Dongjoon Hyun <dhyun@apple.com> Signed-off-by: Dongjoon Hyun <dhyun@apple.com>
2020-01-12 16:18:19 -05:00
ParquetReader Vectorized 286 293 6 54.9 18.2 1.0X
ParquetReader Vectorized -> Row 154 179 57 102.3 9.8 1.9X
[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-30409][SPARK-29173][SQL][TESTS] Use `NoOp` datasource in SQL benchmarks ### What changes were proposed in this pull request? In the PR, I propose to replace `.collect()`, `.count()` and `.foreach(_ => ())` in SQL benchmarks and use the `NoOp` datasource. I added an implicit class to `SqlBasedBenchmark` with the `.noop()` method. It can be used in benchmark like: `ds.noop()`. The last one is unfolded to `ds.write.format("noop").mode(Overwrite).save()`. ### Why are the changes needed? To avoid additional overhead that `collect()` (and other actions) has. For example, `.collect()` has to convert values according to external types and pull data to the driver. This can hide actual performance regressions or improvements of benchmarked operations. ### Does this PR introduce any user-facing change? No ### How was this patch tested? Re-run all modified benchmarks using Amazon EC2. | Item | Description | | ---- | ----| | Region | us-west-2 (Oregon) | | Instance | r3.xlarge (spot instance) | | AMI | ami-06f2f779464715dc5 (ubuntu/images/hvm-ssd/ubuntu-bionic-18.04-amd64-server-20190722.1) | | Java | OpenJDK8/10 | - Run `TPCDSQueryBenchmark` using instructions from the PR #26049 ``` # `spark-tpcds-datagen` needs this. (JDK8) $ git clone https://github.com/apache/spark.git -b branch-2.4 --depth 1 spark-2.4 $ export SPARK_HOME=$PWD $ ./build/mvn clean package -DskipTests # Generate data. (JDK8) $ git clone gitgithub.com:maropu/spark-tpcds-datagen.git $ cd spark-tpcds-datagen/ $ build/mvn clean package $ mkdir -p /data/tpcds $ ./bin/dsdgen --output-location /data/tpcds/s1 // This need `Spark 2.4` ``` - Other benchmarks ran by the script: ``` #!/usr/bin/env python3 import os from sparktestsupport.shellutils import run_cmd benchmarks = [ ['sql/test', 'org.apache.spark.sql.execution.benchmark.AggregateBenchmark'], ['avro/test', 'org.apache.spark.sql.execution.benchmark.AvroReadBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.BloomFilterBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.DataSourceReadBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.DateTimeBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.ExtractBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.FilterPushdownBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.InExpressionBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.IntervalBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.JoinBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.MakeDateTimeBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.MiscBenchmark'], ['hive/test', 'org.apache.spark.sql.execution.benchmark.ObjectHashAggregateExecBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.OrcNestedSchemaPruningBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.OrcV2NestedSchemaPruningBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.ParquetNestedSchemaPruningBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.RangeBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.UDFBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.WideSchemaBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.WideTableBenchmark'], ['hive/test', 'org.apache.spark.sql.hive.orc.OrcReadBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.datasources.csv.CSVBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.datasources.json.JsonBenchmark'] ] print('Set SPARK_GENERATE_BENCHMARK_FILES=1') os.environ['SPARK_GENERATE_BENCHMARK_FILES'] = '1' for b in benchmarks: print("Run benchmark: %s" % b[1]) run_cmd(['build/sbt', '%s:runMain %s' % (b[0], b[1])]) ``` Closes #27078 from MaxGekk/noop-in-benchmarks. Lead-authored-by: Maxim Gekk <max.gekk@gmail.com> Co-authored-by: Maxim Gekk <maxim.gekk@databricks.com> Co-authored-by: Dongjoon Hyun <dhyun@apple.com> Signed-off-by: Dongjoon Hyun <dhyun@apple.com>
2020-01-12 16:18:19 -05:00
OpenJDK 64-Bit Server VM 11.0.5+10-post-Ubuntu-0ubuntu1.118.04 on Linux 4.15.0-1044-aws
[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
Intel(R) Xeon(R) CPU E5-2670 v2 @ 2.50GHz
SQL Single INT Column Scan: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative
------------------------------------------------------------------------------------------------------------------------
[SPARK-30409][SPARK-29173][SQL][TESTS] Use `NoOp` datasource in SQL benchmarks ### What changes were proposed in this pull request? In the PR, I propose to replace `.collect()`, `.count()` and `.foreach(_ => ())` in SQL benchmarks and use the `NoOp` datasource. I added an implicit class to `SqlBasedBenchmark` with the `.noop()` method. It can be used in benchmark like: `ds.noop()`. The last one is unfolded to `ds.write.format("noop").mode(Overwrite).save()`. ### Why are the changes needed? To avoid additional overhead that `collect()` (and other actions) has. For example, `.collect()` has to convert values according to external types and pull data to the driver. This can hide actual performance regressions or improvements of benchmarked operations. ### Does this PR introduce any user-facing change? No ### How was this patch tested? Re-run all modified benchmarks using Amazon EC2. | Item | Description | | ---- | ----| | Region | us-west-2 (Oregon) | | Instance | r3.xlarge (spot instance) | | AMI | ami-06f2f779464715dc5 (ubuntu/images/hvm-ssd/ubuntu-bionic-18.04-amd64-server-20190722.1) | | Java | OpenJDK8/10 | - Run `TPCDSQueryBenchmark` using instructions from the PR #26049 ``` # `spark-tpcds-datagen` needs this. (JDK8) $ git clone https://github.com/apache/spark.git -b branch-2.4 --depth 1 spark-2.4 $ export SPARK_HOME=$PWD $ ./build/mvn clean package -DskipTests # Generate data. (JDK8) $ git clone gitgithub.com:maropu/spark-tpcds-datagen.git $ cd spark-tpcds-datagen/ $ build/mvn clean package $ mkdir -p /data/tpcds $ ./bin/dsdgen --output-location /data/tpcds/s1 // This need `Spark 2.4` ``` - Other benchmarks ran by the script: ``` #!/usr/bin/env python3 import os from sparktestsupport.shellutils import run_cmd benchmarks = [ ['sql/test', 'org.apache.spark.sql.execution.benchmark.AggregateBenchmark'], ['avro/test', 'org.apache.spark.sql.execution.benchmark.AvroReadBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.BloomFilterBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.DataSourceReadBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.DateTimeBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.ExtractBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.FilterPushdownBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.InExpressionBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.IntervalBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.JoinBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.MakeDateTimeBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.MiscBenchmark'], ['hive/test', 'org.apache.spark.sql.execution.benchmark.ObjectHashAggregateExecBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.OrcNestedSchemaPruningBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.OrcV2NestedSchemaPruningBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.ParquetNestedSchemaPruningBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.RangeBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.UDFBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.WideSchemaBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.WideTableBenchmark'], ['hive/test', 'org.apache.spark.sql.hive.orc.OrcReadBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.datasources.csv.CSVBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.datasources.json.JsonBenchmark'] ] print('Set SPARK_GENERATE_BENCHMARK_FILES=1') os.environ['SPARK_GENERATE_BENCHMARK_FILES'] = '1' for b in benchmarks: print("Run benchmark: %s" % b[1]) run_cmd(['build/sbt', '%s:runMain %s' % (b[0], b[1])]) ``` Closes #27078 from MaxGekk/noop-in-benchmarks. Lead-authored-by: Maxim Gekk <max.gekk@gmail.com> Co-authored-by: Maxim Gekk <maxim.gekk@databricks.com> Co-authored-by: Dongjoon Hyun <dhyun@apple.com> Signed-off-by: Dongjoon Hyun <dhyun@apple.com>
2020-01-12 16:18:19 -05:00
SQL CSV 30821 30902 114 0.5 1959.5 1.0X
SQL Json 10935 10944 13 1.4 695.3 2.8X
SQL Parquet Vectorized 203 213 12 77.6 12.9 152.1X
SQL Parquet MR 2334 2351 24 6.7 148.4 13.2X
SQL ORC Vectorized 281 286 4 56.0 17.9 109.6X
SQL ORC MR 1943 2022 112 8.1 123.5 15.9X
[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-30409][SPARK-29173][SQL][TESTS] Use `NoOp` datasource in SQL benchmarks ### What changes were proposed in this pull request? In the PR, I propose to replace `.collect()`, `.count()` and `.foreach(_ => ())` in SQL benchmarks and use the `NoOp` datasource. I added an implicit class to `SqlBasedBenchmark` with the `.noop()` method. It can be used in benchmark like: `ds.noop()`. The last one is unfolded to `ds.write.format("noop").mode(Overwrite).save()`. ### Why are the changes needed? To avoid additional overhead that `collect()` (and other actions) has. For example, `.collect()` has to convert values according to external types and pull data to the driver. This can hide actual performance regressions or improvements of benchmarked operations. ### Does this PR introduce any user-facing change? No ### How was this patch tested? Re-run all modified benchmarks using Amazon EC2. | Item | Description | | ---- | ----| | Region | us-west-2 (Oregon) | | Instance | r3.xlarge (spot instance) | | AMI | ami-06f2f779464715dc5 (ubuntu/images/hvm-ssd/ubuntu-bionic-18.04-amd64-server-20190722.1) | | Java | OpenJDK8/10 | - Run `TPCDSQueryBenchmark` using instructions from the PR #26049 ``` # `spark-tpcds-datagen` needs this. (JDK8) $ git clone https://github.com/apache/spark.git -b branch-2.4 --depth 1 spark-2.4 $ export SPARK_HOME=$PWD $ ./build/mvn clean package -DskipTests # Generate data. (JDK8) $ git clone gitgithub.com:maropu/spark-tpcds-datagen.git $ cd spark-tpcds-datagen/ $ build/mvn clean package $ mkdir -p /data/tpcds $ ./bin/dsdgen --output-location /data/tpcds/s1 // This need `Spark 2.4` ``` - Other benchmarks ran by the script: ``` #!/usr/bin/env python3 import os from sparktestsupport.shellutils import run_cmd benchmarks = [ ['sql/test', 'org.apache.spark.sql.execution.benchmark.AggregateBenchmark'], ['avro/test', 'org.apache.spark.sql.execution.benchmark.AvroReadBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.BloomFilterBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.DataSourceReadBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.DateTimeBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.ExtractBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.FilterPushdownBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.InExpressionBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.IntervalBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.JoinBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.MakeDateTimeBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.MiscBenchmark'], ['hive/test', 'org.apache.spark.sql.execution.benchmark.ObjectHashAggregateExecBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.OrcNestedSchemaPruningBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.OrcV2NestedSchemaPruningBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.ParquetNestedSchemaPruningBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.RangeBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.UDFBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.WideSchemaBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.WideTableBenchmark'], ['hive/test', 'org.apache.spark.sql.hive.orc.OrcReadBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.datasources.csv.CSVBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.datasources.json.JsonBenchmark'] ] print('Set SPARK_GENERATE_BENCHMARK_FILES=1') os.environ['SPARK_GENERATE_BENCHMARK_FILES'] = '1' for b in benchmarks: print("Run benchmark: %s" % b[1]) run_cmd(['build/sbt', '%s:runMain %s' % (b[0], b[1])]) ``` Closes #27078 from MaxGekk/noop-in-benchmarks. Lead-authored-by: Maxim Gekk <max.gekk@gmail.com> Co-authored-by: Maxim Gekk <maxim.gekk@databricks.com> Co-authored-by: Dongjoon Hyun <dhyun@apple.com> Signed-off-by: Dongjoon Hyun <dhyun@apple.com>
2020-01-12 16:18:19 -05:00
OpenJDK 64-Bit Server VM 11.0.5+10-post-Ubuntu-0ubuntu1.118.04 on Linux 4.15.0-1044-aws
[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
Intel(R) Xeon(R) CPU E5-2670 v2 @ 2.50GHz
Parquet Reader Single INT Column Scan: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative
------------------------------------------------------------------------------------------------------------------------
[SPARK-30409][SPARK-29173][SQL][TESTS] Use `NoOp` datasource in SQL benchmarks ### What changes were proposed in this pull request? In the PR, I propose to replace `.collect()`, `.count()` and `.foreach(_ => ())` in SQL benchmarks and use the `NoOp` datasource. I added an implicit class to `SqlBasedBenchmark` with the `.noop()` method. It can be used in benchmark like: `ds.noop()`. The last one is unfolded to `ds.write.format("noop").mode(Overwrite).save()`. ### Why are the changes needed? To avoid additional overhead that `collect()` (and other actions) has. For example, `.collect()` has to convert values according to external types and pull data to the driver. This can hide actual performance regressions or improvements of benchmarked operations. ### Does this PR introduce any user-facing change? No ### How was this patch tested? Re-run all modified benchmarks using Amazon EC2. | Item | Description | | ---- | ----| | Region | us-west-2 (Oregon) | | Instance | r3.xlarge (spot instance) | | AMI | ami-06f2f779464715dc5 (ubuntu/images/hvm-ssd/ubuntu-bionic-18.04-amd64-server-20190722.1) | | Java | OpenJDK8/10 | - Run `TPCDSQueryBenchmark` using instructions from the PR #26049 ``` # `spark-tpcds-datagen` needs this. (JDK8) $ git clone https://github.com/apache/spark.git -b branch-2.4 --depth 1 spark-2.4 $ export SPARK_HOME=$PWD $ ./build/mvn clean package -DskipTests # Generate data. (JDK8) $ git clone gitgithub.com:maropu/spark-tpcds-datagen.git $ cd spark-tpcds-datagen/ $ build/mvn clean package $ mkdir -p /data/tpcds $ ./bin/dsdgen --output-location /data/tpcds/s1 // This need `Spark 2.4` ``` - Other benchmarks ran by the script: ``` #!/usr/bin/env python3 import os from sparktestsupport.shellutils import run_cmd benchmarks = [ ['sql/test', 'org.apache.spark.sql.execution.benchmark.AggregateBenchmark'], ['avro/test', 'org.apache.spark.sql.execution.benchmark.AvroReadBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.BloomFilterBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.DataSourceReadBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.DateTimeBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.ExtractBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.FilterPushdownBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.InExpressionBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.IntervalBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.JoinBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.MakeDateTimeBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.MiscBenchmark'], ['hive/test', 'org.apache.spark.sql.execution.benchmark.ObjectHashAggregateExecBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.OrcNestedSchemaPruningBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.OrcV2NestedSchemaPruningBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.ParquetNestedSchemaPruningBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.RangeBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.UDFBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.WideSchemaBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.WideTableBenchmark'], ['hive/test', 'org.apache.spark.sql.hive.orc.OrcReadBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.datasources.csv.CSVBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.datasources.json.JsonBenchmark'] ] print('Set SPARK_GENERATE_BENCHMARK_FILES=1') os.environ['SPARK_GENERATE_BENCHMARK_FILES'] = '1' for b in benchmarks: print("Run benchmark: %s" % b[1]) run_cmd(['build/sbt', '%s:runMain %s' % (b[0], b[1])]) ``` Closes #27078 from MaxGekk/noop-in-benchmarks. Lead-authored-by: Maxim Gekk <max.gekk@gmail.com> Co-authored-by: Maxim Gekk <maxim.gekk@databricks.com> Co-authored-by: Dongjoon Hyun <dhyun@apple.com> Signed-off-by: Dongjoon Hyun <dhyun@apple.com>
2020-01-12 16:18:19 -05:00
ParquetReader Vectorized 284 291 9 55.5 18.0 1.0X
ParquetReader Vectorized -> Row 277 281 6 56.8 17.6 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-30409][SPARK-29173][SQL][TESTS] Use `NoOp` datasource in SQL benchmarks ### What changes were proposed in this pull request? In the PR, I propose to replace `.collect()`, `.count()` and `.foreach(_ => ())` in SQL benchmarks and use the `NoOp` datasource. I added an implicit class to `SqlBasedBenchmark` with the `.noop()` method. It can be used in benchmark like: `ds.noop()`. The last one is unfolded to `ds.write.format("noop").mode(Overwrite).save()`. ### Why are the changes needed? To avoid additional overhead that `collect()` (and other actions) has. For example, `.collect()` has to convert values according to external types and pull data to the driver. This can hide actual performance regressions or improvements of benchmarked operations. ### Does this PR introduce any user-facing change? No ### How was this patch tested? Re-run all modified benchmarks using Amazon EC2. | Item | Description | | ---- | ----| | Region | us-west-2 (Oregon) | | Instance | r3.xlarge (spot instance) | | AMI | ami-06f2f779464715dc5 (ubuntu/images/hvm-ssd/ubuntu-bionic-18.04-amd64-server-20190722.1) | | Java | OpenJDK8/10 | - Run `TPCDSQueryBenchmark` using instructions from the PR #26049 ``` # `spark-tpcds-datagen` needs this. (JDK8) $ git clone https://github.com/apache/spark.git -b branch-2.4 --depth 1 spark-2.4 $ export SPARK_HOME=$PWD $ ./build/mvn clean package -DskipTests # Generate data. (JDK8) $ git clone gitgithub.com:maropu/spark-tpcds-datagen.git $ cd spark-tpcds-datagen/ $ build/mvn clean package $ mkdir -p /data/tpcds $ ./bin/dsdgen --output-location /data/tpcds/s1 // This need `Spark 2.4` ``` - Other benchmarks ran by the script: ``` #!/usr/bin/env python3 import os from sparktestsupport.shellutils import run_cmd benchmarks = [ ['sql/test', 'org.apache.spark.sql.execution.benchmark.AggregateBenchmark'], ['avro/test', 'org.apache.spark.sql.execution.benchmark.AvroReadBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.BloomFilterBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.DataSourceReadBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.DateTimeBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.ExtractBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.FilterPushdownBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.InExpressionBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.IntervalBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.JoinBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.MakeDateTimeBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.MiscBenchmark'], ['hive/test', 'org.apache.spark.sql.execution.benchmark.ObjectHashAggregateExecBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.OrcNestedSchemaPruningBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.OrcV2NestedSchemaPruningBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.ParquetNestedSchemaPruningBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.RangeBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.UDFBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.WideSchemaBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.WideTableBenchmark'], ['hive/test', 'org.apache.spark.sql.hive.orc.OrcReadBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.datasources.csv.CSVBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.datasources.json.JsonBenchmark'] ] print('Set SPARK_GENERATE_BENCHMARK_FILES=1') os.environ['SPARK_GENERATE_BENCHMARK_FILES'] = '1' for b in benchmarks: print("Run benchmark: %s" % b[1]) run_cmd(['build/sbt', '%s:runMain %s' % (b[0], b[1])]) ``` Closes #27078 from MaxGekk/noop-in-benchmarks. Lead-authored-by: Maxim Gekk <max.gekk@gmail.com> Co-authored-by: Maxim Gekk <maxim.gekk@databricks.com> Co-authored-by: Dongjoon Hyun <dhyun@apple.com> Signed-off-by: Dongjoon Hyun <dhyun@apple.com>
2020-01-12 16:18:19 -05:00
OpenJDK 64-Bit Server VM 11.0.5+10-post-Ubuntu-0ubuntu1.118.04 on Linux 4.15.0-1044-aws
[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
Intel(R) Xeon(R) CPU E5-2670 v2 @ 2.50GHz
SQL Single BIGINT Column Scan: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative
------------------------------------------------------------------------------------------------------------------------
[SPARK-30409][SPARK-29173][SQL][TESTS] Use `NoOp` datasource in SQL benchmarks ### What changes were proposed in this pull request? In the PR, I propose to replace `.collect()`, `.count()` and `.foreach(_ => ())` in SQL benchmarks and use the `NoOp` datasource. I added an implicit class to `SqlBasedBenchmark` with the `.noop()` method. It can be used in benchmark like: `ds.noop()`. The last one is unfolded to `ds.write.format("noop").mode(Overwrite).save()`. ### Why are the changes needed? To avoid additional overhead that `collect()` (and other actions) has. For example, `.collect()` has to convert values according to external types and pull data to the driver. This can hide actual performance regressions or improvements of benchmarked operations. ### Does this PR introduce any user-facing change? No ### How was this patch tested? Re-run all modified benchmarks using Amazon EC2. | Item | Description | | ---- | ----| | Region | us-west-2 (Oregon) | | Instance | r3.xlarge (spot instance) | | AMI | ami-06f2f779464715dc5 (ubuntu/images/hvm-ssd/ubuntu-bionic-18.04-amd64-server-20190722.1) | | Java | OpenJDK8/10 | - Run `TPCDSQueryBenchmark` using instructions from the PR #26049 ``` # `spark-tpcds-datagen` needs this. (JDK8) $ git clone https://github.com/apache/spark.git -b branch-2.4 --depth 1 spark-2.4 $ export SPARK_HOME=$PWD $ ./build/mvn clean package -DskipTests # Generate data. (JDK8) $ git clone gitgithub.com:maropu/spark-tpcds-datagen.git $ cd spark-tpcds-datagen/ $ build/mvn clean package $ mkdir -p /data/tpcds $ ./bin/dsdgen --output-location /data/tpcds/s1 // This need `Spark 2.4` ``` - Other benchmarks ran by the script: ``` #!/usr/bin/env python3 import os from sparktestsupport.shellutils import run_cmd benchmarks = [ ['sql/test', 'org.apache.spark.sql.execution.benchmark.AggregateBenchmark'], ['avro/test', 'org.apache.spark.sql.execution.benchmark.AvroReadBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.BloomFilterBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.DataSourceReadBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.DateTimeBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.ExtractBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.FilterPushdownBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.InExpressionBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.IntervalBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.JoinBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.MakeDateTimeBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.MiscBenchmark'], ['hive/test', 'org.apache.spark.sql.execution.benchmark.ObjectHashAggregateExecBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.OrcNestedSchemaPruningBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.OrcV2NestedSchemaPruningBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.ParquetNestedSchemaPruningBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.RangeBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.UDFBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.WideSchemaBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.WideTableBenchmark'], ['hive/test', 'org.apache.spark.sql.hive.orc.OrcReadBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.datasources.csv.CSVBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.datasources.json.JsonBenchmark'] ] print('Set SPARK_GENERATE_BENCHMARK_FILES=1') os.environ['SPARK_GENERATE_BENCHMARK_FILES'] = '1' for b in benchmarks: print("Run benchmark: %s" % b[1]) run_cmd(['build/sbt', '%s:runMain %s' % (b[0], b[1])]) ``` Closes #27078 from MaxGekk/noop-in-benchmarks. Lead-authored-by: Maxim Gekk <max.gekk@gmail.com> Co-authored-by: Maxim Gekk <maxim.gekk@databricks.com> Co-authored-by: Dongjoon Hyun <dhyun@apple.com> Signed-off-by: Dongjoon Hyun <dhyun@apple.com>
2020-01-12 16:18:19 -05:00
SQL CSV 38264 38306 60 0.4 2432.7 1.0X
SQL Json 14369 14371 3 1.1 913.6 2.7X
SQL Parquet Vectorized 313 319 6 50.3 19.9 122.3X
SQL Parquet MR 2581 2602 30 6.1 164.1 14.8X
SQL ORC Vectorized 423 432 9 37.2 26.9 90.4X
SQL ORC MR 2108 2142 49 7.5 134.0 18.2X
[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-30409][SPARK-29173][SQL][TESTS] Use `NoOp` datasource in SQL benchmarks ### What changes were proposed in this pull request? In the PR, I propose to replace `.collect()`, `.count()` and `.foreach(_ => ())` in SQL benchmarks and use the `NoOp` datasource. I added an implicit class to `SqlBasedBenchmark` with the `.noop()` method. It can be used in benchmark like: `ds.noop()`. The last one is unfolded to `ds.write.format("noop").mode(Overwrite).save()`. ### Why are the changes needed? To avoid additional overhead that `collect()` (and other actions) has. For example, `.collect()` has to convert values according to external types and pull data to the driver. This can hide actual performance regressions or improvements of benchmarked operations. ### Does this PR introduce any user-facing change? No ### How was this patch tested? Re-run all modified benchmarks using Amazon EC2. | Item | Description | | ---- | ----| | Region | us-west-2 (Oregon) | | Instance | r3.xlarge (spot instance) | | AMI | ami-06f2f779464715dc5 (ubuntu/images/hvm-ssd/ubuntu-bionic-18.04-amd64-server-20190722.1) | | Java | OpenJDK8/10 | - Run `TPCDSQueryBenchmark` using instructions from the PR #26049 ``` # `spark-tpcds-datagen` needs this. (JDK8) $ git clone https://github.com/apache/spark.git -b branch-2.4 --depth 1 spark-2.4 $ export SPARK_HOME=$PWD $ ./build/mvn clean package -DskipTests # Generate data. (JDK8) $ git clone gitgithub.com:maropu/spark-tpcds-datagen.git $ cd spark-tpcds-datagen/ $ build/mvn clean package $ mkdir -p /data/tpcds $ ./bin/dsdgen --output-location /data/tpcds/s1 // This need `Spark 2.4` ``` - Other benchmarks ran by the script: ``` #!/usr/bin/env python3 import os from sparktestsupport.shellutils import run_cmd benchmarks = [ ['sql/test', 'org.apache.spark.sql.execution.benchmark.AggregateBenchmark'], ['avro/test', 'org.apache.spark.sql.execution.benchmark.AvroReadBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.BloomFilterBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.DataSourceReadBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.DateTimeBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.ExtractBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.FilterPushdownBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.InExpressionBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.IntervalBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.JoinBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.MakeDateTimeBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.MiscBenchmark'], ['hive/test', 'org.apache.spark.sql.execution.benchmark.ObjectHashAggregateExecBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.OrcNestedSchemaPruningBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.OrcV2NestedSchemaPruningBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.ParquetNestedSchemaPruningBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.RangeBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.UDFBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.WideSchemaBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.WideTableBenchmark'], ['hive/test', 'org.apache.spark.sql.hive.orc.OrcReadBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.datasources.csv.CSVBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.datasources.json.JsonBenchmark'] ] print('Set SPARK_GENERATE_BENCHMARK_FILES=1') os.environ['SPARK_GENERATE_BENCHMARK_FILES'] = '1' for b in benchmarks: print("Run benchmark: %s" % b[1]) run_cmd(['build/sbt', '%s:runMain %s' % (b[0], b[1])]) ``` Closes #27078 from MaxGekk/noop-in-benchmarks. Lead-authored-by: Maxim Gekk <max.gekk@gmail.com> Co-authored-by: Maxim Gekk <maxim.gekk@databricks.com> Co-authored-by: Dongjoon Hyun <dhyun@apple.com> Signed-off-by: Dongjoon Hyun <dhyun@apple.com>
2020-01-12 16:18:19 -05:00
OpenJDK 64-Bit Server VM 11.0.5+10-post-Ubuntu-0ubuntu1.118.04 on Linux 4.15.0-1044-aws
[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
Intel(R) Xeon(R) CPU E5-2670 v2 @ 2.50GHz
Parquet Reader Single BIGINT Column Scan: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative
------------------------------------------------------------------------------------------------------------------------
[SPARK-30409][SPARK-29173][SQL][TESTS] Use `NoOp` datasource in SQL benchmarks ### What changes were proposed in this pull request? In the PR, I propose to replace `.collect()`, `.count()` and `.foreach(_ => ())` in SQL benchmarks and use the `NoOp` datasource. I added an implicit class to `SqlBasedBenchmark` with the `.noop()` method. It can be used in benchmark like: `ds.noop()`. The last one is unfolded to `ds.write.format("noop").mode(Overwrite).save()`. ### Why are the changes needed? To avoid additional overhead that `collect()` (and other actions) has. For example, `.collect()` has to convert values according to external types and pull data to the driver. This can hide actual performance regressions or improvements of benchmarked operations. ### Does this PR introduce any user-facing change? No ### How was this patch tested? Re-run all modified benchmarks using Amazon EC2. | Item | Description | | ---- | ----| | Region | us-west-2 (Oregon) | | Instance | r3.xlarge (spot instance) | | AMI | ami-06f2f779464715dc5 (ubuntu/images/hvm-ssd/ubuntu-bionic-18.04-amd64-server-20190722.1) | | Java | OpenJDK8/10 | - Run `TPCDSQueryBenchmark` using instructions from the PR #26049 ``` # `spark-tpcds-datagen` needs this. (JDK8) $ git clone https://github.com/apache/spark.git -b branch-2.4 --depth 1 spark-2.4 $ export SPARK_HOME=$PWD $ ./build/mvn clean package -DskipTests # Generate data. (JDK8) $ git clone gitgithub.com:maropu/spark-tpcds-datagen.git $ cd spark-tpcds-datagen/ $ build/mvn clean package $ mkdir -p /data/tpcds $ ./bin/dsdgen --output-location /data/tpcds/s1 // This need `Spark 2.4` ``` - Other benchmarks ran by the script: ``` #!/usr/bin/env python3 import os from sparktestsupport.shellutils import run_cmd benchmarks = [ ['sql/test', 'org.apache.spark.sql.execution.benchmark.AggregateBenchmark'], ['avro/test', 'org.apache.spark.sql.execution.benchmark.AvroReadBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.BloomFilterBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.DataSourceReadBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.DateTimeBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.ExtractBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.FilterPushdownBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.InExpressionBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.IntervalBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.JoinBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.MakeDateTimeBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.MiscBenchmark'], ['hive/test', 'org.apache.spark.sql.execution.benchmark.ObjectHashAggregateExecBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.OrcNestedSchemaPruningBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.OrcV2NestedSchemaPruningBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.ParquetNestedSchemaPruningBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.RangeBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.UDFBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.WideSchemaBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.WideTableBenchmark'], ['hive/test', 'org.apache.spark.sql.hive.orc.OrcReadBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.datasources.csv.CSVBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.datasources.json.JsonBenchmark'] ] print('Set SPARK_GENERATE_BENCHMARK_FILES=1') os.environ['SPARK_GENERATE_BENCHMARK_FILES'] = '1' for b in benchmarks: print("Run benchmark: %s" % b[1]) run_cmd(['build/sbt', '%s:runMain %s' % (b[0], b[1])]) ``` Closes #27078 from MaxGekk/noop-in-benchmarks. Lead-authored-by: Maxim Gekk <max.gekk@gmail.com> Co-authored-by: Maxim Gekk <maxim.gekk@databricks.com> Co-authored-by: Dongjoon Hyun <dhyun@apple.com> Signed-off-by: Dongjoon Hyun <dhyun@apple.com>
2020-01-12 16:18:19 -05:00
ParquetReader Vectorized 401 409 8 39.2 25.5 1.0X
ParquetReader Vectorized -> Row 392 400 15 40.2 24.9 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-30409][SPARK-29173][SQL][TESTS] Use `NoOp` datasource in SQL benchmarks ### What changes were proposed in this pull request? In the PR, I propose to replace `.collect()`, `.count()` and `.foreach(_ => ())` in SQL benchmarks and use the `NoOp` datasource. I added an implicit class to `SqlBasedBenchmark` with the `.noop()` method. It can be used in benchmark like: `ds.noop()`. The last one is unfolded to `ds.write.format("noop").mode(Overwrite).save()`. ### Why are the changes needed? To avoid additional overhead that `collect()` (and other actions) has. For example, `.collect()` has to convert values according to external types and pull data to the driver. This can hide actual performance regressions or improvements of benchmarked operations. ### Does this PR introduce any user-facing change? No ### How was this patch tested? Re-run all modified benchmarks using Amazon EC2. | Item | Description | | ---- | ----| | Region | us-west-2 (Oregon) | | Instance | r3.xlarge (spot instance) | | AMI | ami-06f2f779464715dc5 (ubuntu/images/hvm-ssd/ubuntu-bionic-18.04-amd64-server-20190722.1) | | Java | OpenJDK8/10 | - Run `TPCDSQueryBenchmark` using instructions from the PR #26049 ``` # `spark-tpcds-datagen` needs this. (JDK8) $ git clone https://github.com/apache/spark.git -b branch-2.4 --depth 1 spark-2.4 $ export SPARK_HOME=$PWD $ ./build/mvn clean package -DskipTests # Generate data. (JDK8) $ git clone gitgithub.com:maropu/spark-tpcds-datagen.git $ cd spark-tpcds-datagen/ $ build/mvn clean package $ mkdir -p /data/tpcds $ ./bin/dsdgen --output-location /data/tpcds/s1 // This need `Spark 2.4` ``` - Other benchmarks ran by the script: ``` #!/usr/bin/env python3 import os from sparktestsupport.shellutils import run_cmd benchmarks = [ ['sql/test', 'org.apache.spark.sql.execution.benchmark.AggregateBenchmark'], ['avro/test', 'org.apache.spark.sql.execution.benchmark.AvroReadBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.BloomFilterBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.DataSourceReadBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.DateTimeBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.ExtractBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.FilterPushdownBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.InExpressionBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.IntervalBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.JoinBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.MakeDateTimeBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.MiscBenchmark'], ['hive/test', 'org.apache.spark.sql.execution.benchmark.ObjectHashAggregateExecBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.OrcNestedSchemaPruningBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.OrcV2NestedSchemaPruningBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.ParquetNestedSchemaPruningBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.RangeBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.UDFBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.WideSchemaBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.WideTableBenchmark'], ['hive/test', 'org.apache.spark.sql.hive.orc.OrcReadBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.datasources.csv.CSVBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.datasources.json.JsonBenchmark'] ] print('Set SPARK_GENERATE_BENCHMARK_FILES=1') os.environ['SPARK_GENERATE_BENCHMARK_FILES'] = '1' for b in benchmarks: print("Run benchmark: %s" % b[1]) run_cmd(['build/sbt', '%s:runMain %s' % (b[0], b[1])]) ``` Closes #27078 from MaxGekk/noop-in-benchmarks. Lead-authored-by: Maxim Gekk <max.gekk@gmail.com> Co-authored-by: Maxim Gekk <maxim.gekk@databricks.com> Co-authored-by: Dongjoon Hyun <dhyun@apple.com> Signed-off-by: Dongjoon Hyun <dhyun@apple.com>
2020-01-12 16:18:19 -05:00
OpenJDK 64-Bit Server VM 11.0.5+10-post-Ubuntu-0ubuntu1.118.04 on Linux 4.15.0-1044-aws
[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
Intel(R) Xeon(R) CPU E5-2670 v2 @ 2.50GHz
SQL Single FLOAT Column Scan: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative
------------------------------------------------------------------------------------------------------------------------
[SPARK-30409][SPARK-29173][SQL][TESTS] Use `NoOp` datasource in SQL benchmarks ### What changes were proposed in this pull request? In the PR, I propose to replace `.collect()`, `.count()` and `.foreach(_ => ())` in SQL benchmarks and use the `NoOp` datasource. I added an implicit class to `SqlBasedBenchmark` with the `.noop()` method. It can be used in benchmark like: `ds.noop()`. The last one is unfolded to `ds.write.format("noop").mode(Overwrite).save()`. ### Why are the changes needed? To avoid additional overhead that `collect()` (and other actions) has. For example, `.collect()` has to convert values according to external types and pull data to the driver. This can hide actual performance regressions or improvements of benchmarked operations. ### Does this PR introduce any user-facing change? No ### How was this patch tested? Re-run all modified benchmarks using Amazon EC2. | Item | Description | | ---- | ----| | Region | us-west-2 (Oregon) | | Instance | r3.xlarge (spot instance) | | AMI | ami-06f2f779464715dc5 (ubuntu/images/hvm-ssd/ubuntu-bionic-18.04-amd64-server-20190722.1) | | Java | OpenJDK8/10 | - Run `TPCDSQueryBenchmark` using instructions from the PR #26049 ``` # `spark-tpcds-datagen` needs this. (JDK8) $ git clone https://github.com/apache/spark.git -b branch-2.4 --depth 1 spark-2.4 $ export SPARK_HOME=$PWD $ ./build/mvn clean package -DskipTests # Generate data. (JDK8) $ git clone gitgithub.com:maropu/spark-tpcds-datagen.git $ cd spark-tpcds-datagen/ $ build/mvn clean package $ mkdir -p /data/tpcds $ ./bin/dsdgen --output-location /data/tpcds/s1 // This need `Spark 2.4` ``` - Other benchmarks ran by the script: ``` #!/usr/bin/env python3 import os from sparktestsupport.shellutils import run_cmd benchmarks = [ ['sql/test', 'org.apache.spark.sql.execution.benchmark.AggregateBenchmark'], ['avro/test', 'org.apache.spark.sql.execution.benchmark.AvroReadBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.BloomFilterBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.DataSourceReadBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.DateTimeBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.ExtractBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.FilterPushdownBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.InExpressionBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.IntervalBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.JoinBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.MakeDateTimeBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.MiscBenchmark'], ['hive/test', 'org.apache.spark.sql.execution.benchmark.ObjectHashAggregateExecBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.OrcNestedSchemaPruningBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.OrcV2NestedSchemaPruningBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.ParquetNestedSchemaPruningBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.RangeBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.UDFBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.WideSchemaBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.WideTableBenchmark'], ['hive/test', 'org.apache.spark.sql.hive.orc.OrcReadBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.datasources.csv.CSVBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.datasources.json.JsonBenchmark'] ] print('Set SPARK_GENERATE_BENCHMARK_FILES=1') os.environ['SPARK_GENERATE_BENCHMARK_FILES'] = '1' for b in benchmarks: print("Run benchmark: %s" % b[1]) run_cmd(['build/sbt', '%s:runMain %s' % (b[0], b[1])]) ``` Closes #27078 from MaxGekk/noop-in-benchmarks. Lead-authored-by: Maxim Gekk <max.gekk@gmail.com> Co-authored-by: Maxim Gekk <maxim.gekk@databricks.com> Co-authored-by: Dongjoon Hyun <dhyun@apple.com> Signed-off-by: Dongjoon Hyun <dhyun@apple.com>
2020-01-12 16:18:19 -05:00
SQL CSV 36276 36300 34 0.4 2306.4 1.0X
SQL Json 13691 14374 967 1.1 870.4 2.6X
SQL Parquet Vectorized 193 198 5 81.6 12.3 188.2X
SQL Parquet MR 2361 2389 40 6.7 150.1 15.4X
SQL ORC Vectorized 430 434 4 36.6 27.3 84.4X
SQL ORC MR 2037 2072 50 7.7 129.5 17.8X
[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-30409][SPARK-29173][SQL][TESTS] Use `NoOp` datasource in SQL benchmarks ### What changes were proposed in this pull request? In the PR, I propose to replace `.collect()`, `.count()` and `.foreach(_ => ())` in SQL benchmarks and use the `NoOp` datasource. I added an implicit class to `SqlBasedBenchmark` with the `.noop()` method. It can be used in benchmark like: `ds.noop()`. The last one is unfolded to `ds.write.format("noop").mode(Overwrite).save()`. ### Why are the changes needed? To avoid additional overhead that `collect()` (and other actions) has. For example, `.collect()` has to convert values according to external types and pull data to the driver. This can hide actual performance regressions or improvements of benchmarked operations. ### Does this PR introduce any user-facing change? No ### How was this patch tested? Re-run all modified benchmarks using Amazon EC2. | Item | Description | | ---- | ----| | Region | us-west-2 (Oregon) | | Instance | r3.xlarge (spot instance) | | AMI | ami-06f2f779464715dc5 (ubuntu/images/hvm-ssd/ubuntu-bionic-18.04-amd64-server-20190722.1) | | Java | OpenJDK8/10 | - Run `TPCDSQueryBenchmark` using instructions from the PR #26049 ``` # `spark-tpcds-datagen` needs this. (JDK8) $ git clone https://github.com/apache/spark.git -b branch-2.4 --depth 1 spark-2.4 $ export SPARK_HOME=$PWD $ ./build/mvn clean package -DskipTests # Generate data. (JDK8) $ git clone gitgithub.com:maropu/spark-tpcds-datagen.git $ cd spark-tpcds-datagen/ $ build/mvn clean package $ mkdir -p /data/tpcds $ ./bin/dsdgen --output-location /data/tpcds/s1 // This need `Spark 2.4` ``` - Other benchmarks ran by the script: ``` #!/usr/bin/env python3 import os from sparktestsupport.shellutils import run_cmd benchmarks = [ ['sql/test', 'org.apache.spark.sql.execution.benchmark.AggregateBenchmark'], ['avro/test', 'org.apache.spark.sql.execution.benchmark.AvroReadBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.BloomFilterBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.DataSourceReadBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.DateTimeBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.ExtractBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.FilterPushdownBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.InExpressionBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.IntervalBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.JoinBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.MakeDateTimeBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.MiscBenchmark'], ['hive/test', 'org.apache.spark.sql.execution.benchmark.ObjectHashAggregateExecBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.OrcNestedSchemaPruningBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.OrcV2NestedSchemaPruningBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.ParquetNestedSchemaPruningBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.RangeBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.UDFBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.WideSchemaBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.WideTableBenchmark'], ['hive/test', 'org.apache.spark.sql.hive.orc.OrcReadBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.datasources.csv.CSVBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.datasources.json.JsonBenchmark'] ] print('Set SPARK_GENERATE_BENCHMARK_FILES=1') os.environ['SPARK_GENERATE_BENCHMARK_FILES'] = '1' for b in benchmarks: print("Run benchmark: %s" % b[1]) run_cmd(['build/sbt', '%s:runMain %s' % (b[0], b[1])]) ``` Closes #27078 from MaxGekk/noop-in-benchmarks. Lead-authored-by: Maxim Gekk <max.gekk@gmail.com> Co-authored-by: Maxim Gekk <maxim.gekk@databricks.com> Co-authored-by: Dongjoon Hyun <dhyun@apple.com> Signed-off-by: Dongjoon Hyun <dhyun@apple.com>
2020-01-12 16:18:19 -05:00
OpenJDK 64-Bit Server VM 11.0.5+10-post-Ubuntu-0ubuntu1.118.04 on Linux 4.15.0-1044-aws
[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
Intel(R) Xeon(R) CPU E5-2670 v2 @ 2.50GHz
Parquet Reader Single FLOAT Column Scan: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative
------------------------------------------------------------------------------------------------------------------------
[SPARK-30409][SPARK-29173][SQL][TESTS] Use `NoOp` datasource in SQL benchmarks ### What changes were proposed in this pull request? In the PR, I propose to replace `.collect()`, `.count()` and `.foreach(_ => ())` in SQL benchmarks and use the `NoOp` datasource. I added an implicit class to `SqlBasedBenchmark` with the `.noop()` method. It can be used in benchmark like: `ds.noop()`. The last one is unfolded to `ds.write.format("noop").mode(Overwrite).save()`. ### Why are the changes needed? To avoid additional overhead that `collect()` (and other actions) has. For example, `.collect()` has to convert values according to external types and pull data to the driver. This can hide actual performance regressions or improvements of benchmarked operations. ### Does this PR introduce any user-facing change? No ### How was this patch tested? Re-run all modified benchmarks using Amazon EC2. | Item | Description | | ---- | ----| | Region | us-west-2 (Oregon) | | Instance | r3.xlarge (spot instance) | | AMI | ami-06f2f779464715dc5 (ubuntu/images/hvm-ssd/ubuntu-bionic-18.04-amd64-server-20190722.1) | | Java | OpenJDK8/10 | - Run `TPCDSQueryBenchmark` using instructions from the PR #26049 ``` # `spark-tpcds-datagen` needs this. (JDK8) $ git clone https://github.com/apache/spark.git -b branch-2.4 --depth 1 spark-2.4 $ export SPARK_HOME=$PWD $ ./build/mvn clean package -DskipTests # Generate data. (JDK8) $ git clone gitgithub.com:maropu/spark-tpcds-datagen.git $ cd spark-tpcds-datagen/ $ build/mvn clean package $ mkdir -p /data/tpcds $ ./bin/dsdgen --output-location /data/tpcds/s1 // This need `Spark 2.4` ``` - Other benchmarks ran by the script: ``` #!/usr/bin/env python3 import os from sparktestsupport.shellutils import run_cmd benchmarks = [ ['sql/test', 'org.apache.spark.sql.execution.benchmark.AggregateBenchmark'], ['avro/test', 'org.apache.spark.sql.execution.benchmark.AvroReadBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.BloomFilterBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.DataSourceReadBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.DateTimeBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.ExtractBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.FilterPushdownBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.InExpressionBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.IntervalBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.JoinBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.MakeDateTimeBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.MiscBenchmark'], ['hive/test', 'org.apache.spark.sql.execution.benchmark.ObjectHashAggregateExecBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.OrcNestedSchemaPruningBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.OrcV2NestedSchemaPruningBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.ParquetNestedSchemaPruningBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.RangeBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.UDFBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.WideSchemaBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.WideTableBenchmark'], ['hive/test', 'org.apache.spark.sql.hive.orc.OrcReadBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.datasources.csv.CSVBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.datasources.json.JsonBenchmark'] ] print('Set SPARK_GENERATE_BENCHMARK_FILES=1') os.environ['SPARK_GENERATE_BENCHMARK_FILES'] = '1' for b in benchmarks: print("Run benchmark: %s" % b[1]) run_cmd(['build/sbt', '%s:runMain %s' % (b[0], b[1])]) ``` Closes #27078 from MaxGekk/noop-in-benchmarks. Lead-authored-by: Maxim Gekk <max.gekk@gmail.com> Co-authored-by: Maxim Gekk <maxim.gekk@databricks.com> Co-authored-by: Dongjoon Hyun <dhyun@apple.com> Signed-off-by: Dongjoon Hyun <dhyun@apple.com>
2020-01-12 16:18:19 -05:00
ParquetReader Vectorized 277 284 10 56.8 17.6 1.0X
ParquetReader Vectorized -> Row 274 276 4 57.5 17.4 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-30409][SPARK-29173][SQL][TESTS] Use `NoOp` datasource in SQL benchmarks ### What changes were proposed in this pull request? In the PR, I propose to replace `.collect()`, `.count()` and `.foreach(_ => ())` in SQL benchmarks and use the `NoOp` datasource. I added an implicit class to `SqlBasedBenchmark` with the `.noop()` method. It can be used in benchmark like: `ds.noop()`. The last one is unfolded to `ds.write.format("noop").mode(Overwrite).save()`. ### Why are the changes needed? To avoid additional overhead that `collect()` (and other actions) has. For example, `.collect()` has to convert values according to external types and pull data to the driver. This can hide actual performance regressions or improvements of benchmarked operations. ### Does this PR introduce any user-facing change? No ### How was this patch tested? Re-run all modified benchmarks using Amazon EC2. | Item | Description | | ---- | ----| | Region | us-west-2 (Oregon) | | Instance | r3.xlarge (spot instance) | | AMI | ami-06f2f779464715dc5 (ubuntu/images/hvm-ssd/ubuntu-bionic-18.04-amd64-server-20190722.1) | | Java | OpenJDK8/10 | - Run `TPCDSQueryBenchmark` using instructions from the PR #26049 ``` # `spark-tpcds-datagen` needs this. (JDK8) $ git clone https://github.com/apache/spark.git -b branch-2.4 --depth 1 spark-2.4 $ export SPARK_HOME=$PWD $ ./build/mvn clean package -DskipTests # Generate data. (JDK8) $ git clone gitgithub.com:maropu/spark-tpcds-datagen.git $ cd spark-tpcds-datagen/ $ build/mvn clean package $ mkdir -p /data/tpcds $ ./bin/dsdgen --output-location /data/tpcds/s1 // This need `Spark 2.4` ``` - Other benchmarks ran by the script: ``` #!/usr/bin/env python3 import os from sparktestsupport.shellutils import run_cmd benchmarks = [ ['sql/test', 'org.apache.spark.sql.execution.benchmark.AggregateBenchmark'], ['avro/test', 'org.apache.spark.sql.execution.benchmark.AvroReadBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.BloomFilterBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.DataSourceReadBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.DateTimeBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.ExtractBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.FilterPushdownBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.InExpressionBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.IntervalBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.JoinBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.MakeDateTimeBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.MiscBenchmark'], ['hive/test', 'org.apache.spark.sql.execution.benchmark.ObjectHashAggregateExecBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.OrcNestedSchemaPruningBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.OrcV2NestedSchemaPruningBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.ParquetNestedSchemaPruningBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.RangeBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.UDFBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.WideSchemaBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.WideTableBenchmark'], ['hive/test', 'org.apache.spark.sql.hive.orc.OrcReadBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.datasources.csv.CSVBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.datasources.json.JsonBenchmark'] ] print('Set SPARK_GENERATE_BENCHMARK_FILES=1') os.environ['SPARK_GENERATE_BENCHMARK_FILES'] = '1' for b in benchmarks: print("Run benchmark: %s" % b[1]) run_cmd(['build/sbt', '%s:runMain %s' % (b[0], b[1])]) ``` Closes #27078 from MaxGekk/noop-in-benchmarks. Lead-authored-by: Maxim Gekk <max.gekk@gmail.com> Co-authored-by: Maxim Gekk <maxim.gekk@databricks.com> Co-authored-by: Dongjoon Hyun <dhyun@apple.com> Signed-off-by: Dongjoon Hyun <dhyun@apple.com>
2020-01-12 16:18:19 -05:00
OpenJDK 64-Bit Server VM 11.0.5+10-post-Ubuntu-0ubuntu1.118.04 on Linux 4.15.0-1044-aws
[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
Intel(R) Xeon(R) CPU E5-2670 v2 @ 2.50GHz
SQL Single DOUBLE Column Scan: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative
------------------------------------------------------------------------------------------------------------------------
[SPARK-30409][SPARK-29173][SQL][TESTS] Use `NoOp` datasource in SQL benchmarks ### What changes were proposed in this pull request? In the PR, I propose to replace `.collect()`, `.count()` and `.foreach(_ => ())` in SQL benchmarks and use the `NoOp` datasource. I added an implicit class to `SqlBasedBenchmark` with the `.noop()` method. It can be used in benchmark like: `ds.noop()`. The last one is unfolded to `ds.write.format("noop").mode(Overwrite).save()`. ### Why are the changes needed? To avoid additional overhead that `collect()` (and other actions) has. For example, `.collect()` has to convert values according to external types and pull data to the driver. This can hide actual performance regressions or improvements of benchmarked operations. ### Does this PR introduce any user-facing change? No ### How was this patch tested? Re-run all modified benchmarks using Amazon EC2. | Item | Description | | ---- | ----| | Region | us-west-2 (Oregon) | | Instance | r3.xlarge (spot instance) | | AMI | ami-06f2f779464715dc5 (ubuntu/images/hvm-ssd/ubuntu-bionic-18.04-amd64-server-20190722.1) | | Java | OpenJDK8/10 | - Run `TPCDSQueryBenchmark` using instructions from the PR #26049 ``` # `spark-tpcds-datagen` needs this. (JDK8) $ git clone https://github.com/apache/spark.git -b branch-2.4 --depth 1 spark-2.4 $ export SPARK_HOME=$PWD $ ./build/mvn clean package -DskipTests # Generate data. (JDK8) $ git clone gitgithub.com:maropu/spark-tpcds-datagen.git $ cd spark-tpcds-datagen/ $ build/mvn clean package $ mkdir -p /data/tpcds $ ./bin/dsdgen --output-location /data/tpcds/s1 // This need `Spark 2.4` ``` - Other benchmarks ran by the script: ``` #!/usr/bin/env python3 import os from sparktestsupport.shellutils import run_cmd benchmarks = [ ['sql/test', 'org.apache.spark.sql.execution.benchmark.AggregateBenchmark'], ['avro/test', 'org.apache.spark.sql.execution.benchmark.AvroReadBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.BloomFilterBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.DataSourceReadBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.DateTimeBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.ExtractBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.FilterPushdownBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.InExpressionBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.IntervalBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.JoinBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.MakeDateTimeBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.MiscBenchmark'], ['hive/test', 'org.apache.spark.sql.execution.benchmark.ObjectHashAggregateExecBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.OrcNestedSchemaPruningBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.OrcV2NestedSchemaPruningBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.ParquetNestedSchemaPruningBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.RangeBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.UDFBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.WideSchemaBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.WideTableBenchmark'], ['hive/test', 'org.apache.spark.sql.hive.orc.OrcReadBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.datasources.csv.CSVBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.datasources.json.JsonBenchmark'] ] print('Set SPARK_GENERATE_BENCHMARK_FILES=1') os.environ['SPARK_GENERATE_BENCHMARK_FILES'] = '1' for b in benchmarks: print("Run benchmark: %s" % b[1]) run_cmd(['build/sbt', '%s:runMain %s' % (b[0], b[1])]) ``` Closes #27078 from MaxGekk/noop-in-benchmarks. Lead-authored-by: Maxim Gekk <max.gekk@gmail.com> Co-authored-by: Maxim Gekk <maxim.gekk@databricks.com> Co-authored-by: Dongjoon Hyun <dhyun@apple.com> Signed-off-by: Dongjoon Hyun <dhyun@apple.com>
2020-01-12 16:18:19 -05:00
SQL CSV 39757 39761 5 0.4 2527.7 1.0X
SQL Json 20049 20052 5 0.8 1274.7 2.0X
SQL Parquet Vectorized 310 318 10 50.7 19.7 128.3X
SQL Parquet MR 2535 2571 52 6.2 161.2 15.7X
SQL ORC Vectorized 537 543 8 29.3 34.1 74.1X
SQL ORC MR 2132 2161 41 7.4 135.6 18.6X
[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-30409][SPARK-29173][SQL][TESTS] Use `NoOp` datasource in SQL benchmarks ### What changes were proposed in this pull request? In the PR, I propose to replace `.collect()`, `.count()` and `.foreach(_ => ())` in SQL benchmarks and use the `NoOp` datasource. I added an implicit class to `SqlBasedBenchmark` with the `.noop()` method. It can be used in benchmark like: `ds.noop()`. The last one is unfolded to `ds.write.format("noop").mode(Overwrite).save()`. ### Why are the changes needed? To avoid additional overhead that `collect()` (and other actions) has. For example, `.collect()` has to convert values according to external types and pull data to the driver. This can hide actual performance regressions or improvements of benchmarked operations. ### Does this PR introduce any user-facing change? No ### How was this patch tested? Re-run all modified benchmarks using Amazon EC2. | Item | Description | | ---- | ----| | Region | us-west-2 (Oregon) | | Instance | r3.xlarge (spot instance) | | AMI | ami-06f2f779464715dc5 (ubuntu/images/hvm-ssd/ubuntu-bionic-18.04-amd64-server-20190722.1) | | Java | OpenJDK8/10 | - Run `TPCDSQueryBenchmark` using instructions from the PR #26049 ``` # `spark-tpcds-datagen` needs this. (JDK8) $ git clone https://github.com/apache/spark.git -b branch-2.4 --depth 1 spark-2.4 $ export SPARK_HOME=$PWD $ ./build/mvn clean package -DskipTests # Generate data. (JDK8) $ git clone gitgithub.com:maropu/spark-tpcds-datagen.git $ cd spark-tpcds-datagen/ $ build/mvn clean package $ mkdir -p /data/tpcds $ ./bin/dsdgen --output-location /data/tpcds/s1 // This need `Spark 2.4` ``` - Other benchmarks ran by the script: ``` #!/usr/bin/env python3 import os from sparktestsupport.shellutils import run_cmd benchmarks = [ ['sql/test', 'org.apache.spark.sql.execution.benchmark.AggregateBenchmark'], ['avro/test', 'org.apache.spark.sql.execution.benchmark.AvroReadBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.BloomFilterBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.DataSourceReadBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.DateTimeBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.ExtractBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.FilterPushdownBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.InExpressionBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.IntervalBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.JoinBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.MakeDateTimeBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.MiscBenchmark'], ['hive/test', 'org.apache.spark.sql.execution.benchmark.ObjectHashAggregateExecBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.OrcNestedSchemaPruningBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.OrcV2NestedSchemaPruningBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.ParquetNestedSchemaPruningBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.RangeBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.UDFBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.WideSchemaBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.WideTableBenchmark'], ['hive/test', 'org.apache.spark.sql.hive.orc.OrcReadBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.datasources.csv.CSVBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.datasources.json.JsonBenchmark'] ] print('Set SPARK_GENERATE_BENCHMARK_FILES=1') os.environ['SPARK_GENERATE_BENCHMARK_FILES'] = '1' for b in benchmarks: print("Run benchmark: %s" % b[1]) run_cmd(['build/sbt', '%s:runMain %s' % (b[0], b[1])]) ``` Closes #27078 from MaxGekk/noop-in-benchmarks. Lead-authored-by: Maxim Gekk <max.gekk@gmail.com> Co-authored-by: Maxim Gekk <maxim.gekk@databricks.com> Co-authored-by: Dongjoon Hyun <dhyun@apple.com> Signed-off-by: Dongjoon Hyun <dhyun@apple.com>
2020-01-12 16:18:19 -05:00
OpenJDK 64-Bit Server VM 11.0.5+10-post-Ubuntu-0ubuntu1.118.04 on Linux 4.15.0-1044-aws
[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
Intel(R) Xeon(R) CPU E5-2670 v2 @ 2.50GHz
Parquet Reader Single DOUBLE Column Scan: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative
------------------------------------------------------------------------------------------------------------------------
[SPARK-30409][SPARK-29173][SQL][TESTS] Use `NoOp` datasource in SQL benchmarks ### What changes were proposed in this pull request? In the PR, I propose to replace `.collect()`, `.count()` and `.foreach(_ => ())` in SQL benchmarks and use the `NoOp` datasource. I added an implicit class to `SqlBasedBenchmark` with the `.noop()` method. It can be used in benchmark like: `ds.noop()`. The last one is unfolded to `ds.write.format("noop").mode(Overwrite).save()`. ### Why are the changes needed? To avoid additional overhead that `collect()` (and other actions) has. For example, `.collect()` has to convert values according to external types and pull data to the driver. This can hide actual performance regressions or improvements of benchmarked operations. ### Does this PR introduce any user-facing change? No ### How was this patch tested? Re-run all modified benchmarks using Amazon EC2. | Item | Description | | ---- | ----| | Region | us-west-2 (Oregon) | | Instance | r3.xlarge (spot instance) | | AMI | ami-06f2f779464715dc5 (ubuntu/images/hvm-ssd/ubuntu-bionic-18.04-amd64-server-20190722.1) | | Java | OpenJDK8/10 | - Run `TPCDSQueryBenchmark` using instructions from the PR #26049 ``` # `spark-tpcds-datagen` needs this. (JDK8) $ git clone https://github.com/apache/spark.git -b branch-2.4 --depth 1 spark-2.4 $ export SPARK_HOME=$PWD $ ./build/mvn clean package -DskipTests # Generate data. (JDK8) $ git clone gitgithub.com:maropu/spark-tpcds-datagen.git $ cd spark-tpcds-datagen/ $ build/mvn clean package $ mkdir -p /data/tpcds $ ./bin/dsdgen --output-location /data/tpcds/s1 // This need `Spark 2.4` ``` - Other benchmarks ran by the script: ``` #!/usr/bin/env python3 import os from sparktestsupport.shellutils import run_cmd benchmarks = [ ['sql/test', 'org.apache.spark.sql.execution.benchmark.AggregateBenchmark'], ['avro/test', 'org.apache.spark.sql.execution.benchmark.AvroReadBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.BloomFilterBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.DataSourceReadBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.DateTimeBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.ExtractBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.FilterPushdownBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.InExpressionBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.IntervalBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.JoinBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.MakeDateTimeBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.MiscBenchmark'], ['hive/test', 'org.apache.spark.sql.execution.benchmark.ObjectHashAggregateExecBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.OrcNestedSchemaPruningBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.OrcV2NestedSchemaPruningBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.ParquetNestedSchemaPruningBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.RangeBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.UDFBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.WideSchemaBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.WideTableBenchmark'], ['hive/test', 'org.apache.spark.sql.hive.orc.OrcReadBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.datasources.csv.CSVBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.datasources.json.JsonBenchmark'] ] print('Set SPARK_GENERATE_BENCHMARK_FILES=1') os.environ['SPARK_GENERATE_BENCHMARK_FILES'] = '1' for b in benchmarks: print("Run benchmark: %s" % b[1]) run_cmd(['build/sbt', '%s:runMain %s' % (b[0], b[1])]) ``` Closes #27078 from MaxGekk/noop-in-benchmarks. Lead-authored-by: Maxim Gekk <max.gekk@gmail.com> Co-authored-by: Maxim Gekk <maxim.gekk@databricks.com> Co-authored-by: Dongjoon Hyun <dhyun@apple.com> Signed-off-by: Dongjoon Hyun <dhyun@apple.com>
2020-01-12 16:18:19 -05:00
ParquetReader Vectorized 390 394 5 40.3 24.8 1.0X
ParquetReader Vectorized -> Row 389 391 5 40.5 24.7 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
================================================================================================
Int and String Scan
================================================================================================
[SPARK-30409][SPARK-29173][SQL][TESTS] Use `NoOp` datasource in SQL benchmarks ### What changes were proposed in this pull request? In the PR, I propose to replace `.collect()`, `.count()` and `.foreach(_ => ())` in SQL benchmarks and use the `NoOp` datasource. I added an implicit class to `SqlBasedBenchmark` with the `.noop()` method. It can be used in benchmark like: `ds.noop()`. The last one is unfolded to `ds.write.format("noop").mode(Overwrite).save()`. ### Why are the changes needed? To avoid additional overhead that `collect()` (and other actions) has. For example, `.collect()` has to convert values according to external types and pull data to the driver. This can hide actual performance regressions or improvements of benchmarked operations. ### Does this PR introduce any user-facing change? No ### How was this patch tested? Re-run all modified benchmarks using Amazon EC2. | Item | Description | | ---- | ----| | Region | us-west-2 (Oregon) | | Instance | r3.xlarge (spot instance) | | AMI | ami-06f2f779464715dc5 (ubuntu/images/hvm-ssd/ubuntu-bionic-18.04-amd64-server-20190722.1) | | Java | OpenJDK8/10 | - Run `TPCDSQueryBenchmark` using instructions from the PR #26049 ``` # `spark-tpcds-datagen` needs this. (JDK8) $ git clone https://github.com/apache/spark.git -b branch-2.4 --depth 1 spark-2.4 $ export SPARK_HOME=$PWD $ ./build/mvn clean package -DskipTests # Generate data. (JDK8) $ git clone gitgithub.com:maropu/spark-tpcds-datagen.git $ cd spark-tpcds-datagen/ $ build/mvn clean package $ mkdir -p /data/tpcds $ ./bin/dsdgen --output-location /data/tpcds/s1 // This need `Spark 2.4` ``` - Other benchmarks ran by the script: ``` #!/usr/bin/env python3 import os from sparktestsupport.shellutils import run_cmd benchmarks = [ ['sql/test', 'org.apache.spark.sql.execution.benchmark.AggregateBenchmark'], ['avro/test', 'org.apache.spark.sql.execution.benchmark.AvroReadBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.BloomFilterBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.DataSourceReadBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.DateTimeBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.ExtractBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.FilterPushdownBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.InExpressionBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.IntervalBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.JoinBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.MakeDateTimeBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.MiscBenchmark'], ['hive/test', 'org.apache.spark.sql.execution.benchmark.ObjectHashAggregateExecBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.OrcNestedSchemaPruningBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.OrcV2NestedSchemaPruningBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.ParquetNestedSchemaPruningBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.RangeBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.UDFBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.WideSchemaBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.WideTableBenchmark'], ['hive/test', 'org.apache.spark.sql.hive.orc.OrcReadBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.datasources.csv.CSVBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.datasources.json.JsonBenchmark'] ] print('Set SPARK_GENERATE_BENCHMARK_FILES=1') os.environ['SPARK_GENERATE_BENCHMARK_FILES'] = '1' for b in benchmarks: print("Run benchmark: %s" % b[1]) run_cmd(['build/sbt', '%s:runMain %s' % (b[0], b[1])]) ``` Closes #27078 from MaxGekk/noop-in-benchmarks. Lead-authored-by: Maxim Gekk <max.gekk@gmail.com> Co-authored-by: Maxim Gekk <maxim.gekk@databricks.com> Co-authored-by: Dongjoon Hyun <dhyun@apple.com> Signed-off-by: Dongjoon Hyun <dhyun@apple.com>
2020-01-12 16:18:19 -05:00
OpenJDK 64-Bit Server VM 11.0.5+10-post-Ubuntu-0ubuntu1.118.04 on Linux 4.15.0-1044-aws
[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
Intel(R) Xeon(R) CPU E5-2670 v2 @ 2.50GHz
Int and String Scan: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative
------------------------------------------------------------------------------------------------------------------------
[SPARK-30409][SPARK-29173][SQL][TESTS] Use `NoOp` datasource in SQL benchmarks ### What changes were proposed in this pull request? In the PR, I propose to replace `.collect()`, `.count()` and `.foreach(_ => ())` in SQL benchmarks and use the `NoOp` datasource. I added an implicit class to `SqlBasedBenchmark` with the `.noop()` method. It can be used in benchmark like: `ds.noop()`. The last one is unfolded to `ds.write.format("noop").mode(Overwrite).save()`. ### Why are the changes needed? To avoid additional overhead that `collect()` (and other actions) has. For example, `.collect()` has to convert values according to external types and pull data to the driver. This can hide actual performance regressions or improvements of benchmarked operations. ### Does this PR introduce any user-facing change? No ### How was this patch tested? Re-run all modified benchmarks using Amazon EC2. | Item | Description | | ---- | ----| | Region | us-west-2 (Oregon) | | Instance | r3.xlarge (spot instance) | | AMI | ami-06f2f779464715dc5 (ubuntu/images/hvm-ssd/ubuntu-bionic-18.04-amd64-server-20190722.1) | | Java | OpenJDK8/10 | - Run `TPCDSQueryBenchmark` using instructions from the PR #26049 ``` # `spark-tpcds-datagen` needs this. (JDK8) $ git clone https://github.com/apache/spark.git -b branch-2.4 --depth 1 spark-2.4 $ export SPARK_HOME=$PWD $ ./build/mvn clean package -DskipTests # Generate data. (JDK8) $ git clone gitgithub.com:maropu/spark-tpcds-datagen.git $ cd spark-tpcds-datagen/ $ build/mvn clean package $ mkdir -p /data/tpcds $ ./bin/dsdgen --output-location /data/tpcds/s1 // This need `Spark 2.4` ``` - Other benchmarks ran by the script: ``` #!/usr/bin/env python3 import os from sparktestsupport.shellutils import run_cmd benchmarks = [ ['sql/test', 'org.apache.spark.sql.execution.benchmark.AggregateBenchmark'], ['avro/test', 'org.apache.spark.sql.execution.benchmark.AvroReadBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.BloomFilterBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.DataSourceReadBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.DateTimeBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.ExtractBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.FilterPushdownBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.InExpressionBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.IntervalBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.JoinBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.MakeDateTimeBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.MiscBenchmark'], ['hive/test', 'org.apache.spark.sql.execution.benchmark.ObjectHashAggregateExecBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.OrcNestedSchemaPruningBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.OrcV2NestedSchemaPruningBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.ParquetNestedSchemaPruningBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.RangeBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.UDFBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.WideSchemaBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.WideTableBenchmark'], ['hive/test', 'org.apache.spark.sql.hive.orc.OrcReadBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.datasources.csv.CSVBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.datasources.json.JsonBenchmark'] ] print('Set SPARK_GENERATE_BENCHMARK_FILES=1') os.environ['SPARK_GENERATE_BENCHMARK_FILES'] = '1' for b in benchmarks: print("Run benchmark: %s" % b[1]) run_cmd(['build/sbt', '%s:runMain %s' % (b[0], b[1])]) ``` Closes #27078 from MaxGekk/noop-in-benchmarks. Lead-authored-by: Maxim Gekk <max.gekk@gmail.com> Co-authored-by: Maxim Gekk <maxim.gekk@databricks.com> Co-authored-by: Dongjoon Hyun <dhyun@apple.com> Signed-off-by: Dongjoon Hyun <dhyun@apple.com>
2020-01-12 16:18:19 -05:00
SQL CSV 27215 27240 34 0.4 2595.5 1.0X
SQL Json 12713 12783 100 0.8 1212.4 2.1X
SQL Parquet Vectorized 2265 2269 5 4.6 216.0 12.0X
SQL Parquet MR 4477 4544 95 2.3 426.9 6.1X
SQL ORC Vectorized 2388 2404 23 4.4 227.7 11.4X
SQL ORC MR 4295 4305 15 2.4 409.6 6.3X
[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
================================================================================================
Repeated String Scan
================================================================================================
[SPARK-30409][SPARK-29173][SQL][TESTS] Use `NoOp` datasource in SQL benchmarks ### What changes were proposed in this pull request? In the PR, I propose to replace `.collect()`, `.count()` and `.foreach(_ => ())` in SQL benchmarks and use the `NoOp` datasource. I added an implicit class to `SqlBasedBenchmark` with the `.noop()` method. It can be used in benchmark like: `ds.noop()`. The last one is unfolded to `ds.write.format("noop").mode(Overwrite).save()`. ### Why are the changes needed? To avoid additional overhead that `collect()` (and other actions) has. For example, `.collect()` has to convert values according to external types and pull data to the driver. This can hide actual performance regressions or improvements of benchmarked operations. ### Does this PR introduce any user-facing change? No ### How was this patch tested? Re-run all modified benchmarks using Amazon EC2. | Item | Description | | ---- | ----| | Region | us-west-2 (Oregon) | | Instance | r3.xlarge (spot instance) | | AMI | ami-06f2f779464715dc5 (ubuntu/images/hvm-ssd/ubuntu-bionic-18.04-amd64-server-20190722.1) | | Java | OpenJDK8/10 | - Run `TPCDSQueryBenchmark` using instructions from the PR #26049 ``` # `spark-tpcds-datagen` needs this. (JDK8) $ git clone https://github.com/apache/spark.git -b branch-2.4 --depth 1 spark-2.4 $ export SPARK_HOME=$PWD $ ./build/mvn clean package -DskipTests # Generate data. (JDK8) $ git clone gitgithub.com:maropu/spark-tpcds-datagen.git $ cd spark-tpcds-datagen/ $ build/mvn clean package $ mkdir -p /data/tpcds $ ./bin/dsdgen --output-location /data/tpcds/s1 // This need `Spark 2.4` ``` - Other benchmarks ran by the script: ``` #!/usr/bin/env python3 import os from sparktestsupport.shellutils import run_cmd benchmarks = [ ['sql/test', 'org.apache.spark.sql.execution.benchmark.AggregateBenchmark'], ['avro/test', 'org.apache.spark.sql.execution.benchmark.AvroReadBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.BloomFilterBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.DataSourceReadBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.DateTimeBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.ExtractBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.FilterPushdownBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.InExpressionBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.IntervalBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.JoinBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.MakeDateTimeBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.MiscBenchmark'], ['hive/test', 'org.apache.spark.sql.execution.benchmark.ObjectHashAggregateExecBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.OrcNestedSchemaPruningBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.OrcV2NestedSchemaPruningBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.ParquetNestedSchemaPruningBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.RangeBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.UDFBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.WideSchemaBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.WideTableBenchmark'], ['hive/test', 'org.apache.spark.sql.hive.orc.OrcReadBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.datasources.csv.CSVBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.datasources.json.JsonBenchmark'] ] print('Set SPARK_GENERATE_BENCHMARK_FILES=1') os.environ['SPARK_GENERATE_BENCHMARK_FILES'] = '1' for b in benchmarks: print("Run benchmark: %s" % b[1]) run_cmd(['build/sbt', '%s:runMain %s' % (b[0], b[1])]) ``` Closes #27078 from MaxGekk/noop-in-benchmarks. Lead-authored-by: Maxim Gekk <max.gekk@gmail.com> Co-authored-by: Maxim Gekk <maxim.gekk@databricks.com> Co-authored-by: Dongjoon Hyun <dhyun@apple.com> Signed-off-by: Dongjoon Hyun <dhyun@apple.com>
2020-01-12 16:18:19 -05:00
OpenJDK 64-Bit Server VM 11.0.5+10-post-Ubuntu-0ubuntu1.118.04 on Linux 4.15.0-1044-aws
[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
Intel(R) Xeon(R) CPU E5-2670 v2 @ 2.50GHz
Repeated String: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative
------------------------------------------------------------------------------------------------------------------------
[SPARK-30409][SPARK-29173][SQL][TESTS] Use `NoOp` datasource in SQL benchmarks ### What changes were proposed in this pull request? In the PR, I propose to replace `.collect()`, `.count()` and `.foreach(_ => ())` in SQL benchmarks and use the `NoOp` datasource. I added an implicit class to `SqlBasedBenchmark` with the `.noop()` method. It can be used in benchmark like: `ds.noop()`. The last one is unfolded to `ds.write.format("noop").mode(Overwrite).save()`. ### Why are the changes needed? To avoid additional overhead that `collect()` (and other actions) has. For example, `.collect()` has to convert values according to external types and pull data to the driver. This can hide actual performance regressions or improvements of benchmarked operations. ### Does this PR introduce any user-facing change? No ### How was this patch tested? Re-run all modified benchmarks using Amazon EC2. | Item | Description | | ---- | ----| | Region | us-west-2 (Oregon) | | Instance | r3.xlarge (spot instance) | | AMI | ami-06f2f779464715dc5 (ubuntu/images/hvm-ssd/ubuntu-bionic-18.04-amd64-server-20190722.1) | | Java | OpenJDK8/10 | - Run `TPCDSQueryBenchmark` using instructions from the PR #26049 ``` # `spark-tpcds-datagen` needs this. (JDK8) $ git clone https://github.com/apache/spark.git -b branch-2.4 --depth 1 spark-2.4 $ export SPARK_HOME=$PWD $ ./build/mvn clean package -DskipTests # Generate data. (JDK8) $ git clone gitgithub.com:maropu/spark-tpcds-datagen.git $ cd spark-tpcds-datagen/ $ build/mvn clean package $ mkdir -p /data/tpcds $ ./bin/dsdgen --output-location /data/tpcds/s1 // This need `Spark 2.4` ``` - Other benchmarks ran by the script: ``` #!/usr/bin/env python3 import os from sparktestsupport.shellutils import run_cmd benchmarks = [ ['sql/test', 'org.apache.spark.sql.execution.benchmark.AggregateBenchmark'], ['avro/test', 'org.apache.spark.sql.execution.benchmark.AvroReadBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.BloomFilterBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.DataSourceReadBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.DateTimeBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.ExtractBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.FilterPushdownBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.InExpressionBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.IntervalBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.JoinBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.MakeDateTimeBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.MiscBenchmark'], ['hive/test', 'org.apache.spark.sql.execution.benchmark.ObjectHashAggregateExecBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.OrcNestedSchemaPruningBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.OrcV2NestedSchemaPruningBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.ParquetNestedSchemaPruningBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.RangeBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.UDFBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.WideSchemaBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.WideTableBenchmark'], ['hive/test', 'org.apache.spark.sql.hive.orc.OrcReadBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.datasources.csv.CSVBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.datasources.json.JsonBenchmark'] ] print('Set SPARK_GENERATE_BENCHMARK_FILES=1') os.environ['SPARK_GENERATE_BENCHMARK_FILES'] = '1' for b in benchmarks: print("Run benchmark: %s" % b[1]) run_cmd(['build/sbt', '%s:runMain %s' % (b[0], b[1])]) ``` Closes #27078 from MaxGekk/noop-in-benchmarks. Lead-authored-by: Maxim Gekk <max.gekk@gmail.com> Co-authored-by: Maxim Gekk <maxim.gekk@databricks.com> Co-authored-by: Dongjoon Hyun <dhyun@apple.com> Signed-off-by: Dongjoon Hyun <dhyun@apple.com>
2020-01-12 16:18:19 -05:00
SQL CSV 17544 17580 51 0.6 1673.1 1.0X
SQL Json 8277 8328 71 1.3 789.4 2.1X
SQL Parquet Vectorized 674 682 7 15.6 64.3 26.0X
SQL Parquet MR 1960 1972 17 5.3 187.0 8.9X
SQL ORC Vectorized 551 558 11 19.0 52.6 31.8X
SQL ORC MR 2047 2052 6 5.1 195.2 8.6X
[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
================================================================================================
Partitioned Table Scan
================================================================================================
[SPARK-30409][SPARK-29173][SQL][TESTS] Use `NoOp` datasource in SQL benchmarks ### What changes were proposed in this pull request? In the PR, I propose to replace `.collect()`, `.count()` and `.foreach(_ => ())` in SQL benchmarks and use the `NoOp` datasource. I added an implicit class to `SqlBasedBenchmark` with the `.noop()` method. It can be used in benchmark like: `ds.noop()`. The last one is unfolded to `ds.write.format("noop").mode(Overwrite).save()`. ### Why are the changes needed? To avoid additional overhead that `collect()` (and other actions) has. For example, `.collect()` has to convert values according to external types and pull data to the driver. This can hide actual performance regressions or improvements of benchmarked operations. ### Does this PR introduce any user-facing change? No ### How was this patch tested? Re-run all modified benchmarks using Amazon EC2. | Item | Description | | ---- | ----| | Region | us-west-2 (Oregon) | | Instance | r3.xlarge (spot instance) | | AMI | ami-06f2f779464715dc5 (ubuntu/images/hvm-ssd/ubuntu-bionic-18.04-amd64-server-20190722.1) | | Java | OpenJDK8/10 | - Run `TPCDSQueryBenchmark` using instructions from the PR #26049 ``` # `spark-tpcds-datagen` needs this. (JDK8) $ git clone https://github.com/apache/spark.git -b branch-2.4 --depth 1 spark-2.4 $ export SPARK_HOME=$PWD $ ./build/mvn clean package -DskipTests # Generate data. (JDK8) $ git clone gitgithub.com:maropu/spark-tpcds-datagen.git $ cd spark-tpcds-datagen/ $ build/mvn clean package $ mkdir -p /data/tpcds $ ./bin/dsdgen --output-location /data/tpcds/s1 // This need `Spark 2.4` ``` - Other benchmarks ran by the script: ``` #!/usr/bin/env python3 import os from sparktestsupport.shellutils import run_cmd benchmarks = [ ['sql/test', 'org.apache.spark.sql.execution.benchmark.AggregateBenchmark'], ['avro/test', 'org.apache.spark.sql.execution.benchmark.AvroReadBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.BloomFilterBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.DataSourceReadBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.DateTimeBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.ExtractBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.FilterPushdownBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.InExpressionBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.IntervalBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.JoinBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.MakeDateTimeBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.MiscBenchmark'], ['hive/test', 'org.apache.spark.sql.execution.benchmark.ObjectHashAggregateExecBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.OrcNestedSchemaPruningBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.OrcV2NestedSchemaPruningBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.ParquetNestedSchemaPruningBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.RangeBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.UDFBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.WideSchemaBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.WideTableBenchmark'], ['hive/test', 'org.apache.spark.sql.hive.orc.OrcReadBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.datasources.csv.CSVBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.datasources.json.JsonBenchmark'] ] print('Set SPARK_GENERATE_BENCHMARK_FILES=1') os.environ['SPARK_GENERATE_BENCHMARK_FILES'] = '1' for b in benchmarks: print("Run benchmark: %s" % b[1]) run_cmd(['build/sbt', '%s:runMain %s' % (b[0], b[1])]) ``` Closes #27078 from MaxGekk/noop-in-benchmarks. Lead-authored-by: Maxim Gekk <max.gekk@gmail.com> Co-authored-by: Maxim Gekk <maxim.gekk@databricks.com> Co-authored-by: Dongjoon Hyun <dhyun@apple.com> Signed-off-by: Dongjoon Hyun <dhyun@apple.com>
2020-01-12 16:18:19 -05:00
OpenJDK 64-Bit Server VM 11.0.5+10-post-Ubuntu-0ubuntu1.118.04 on Linux 4.15.0-1044-aws
[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
Intel(R) Xeon(R) CPU E5-2670 v2 @ 2.50GHz
Partitioned Table: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative
------------------------------------------------------------------------------------------------------------------------
[SPARK-30409][SPARK-29173][SQL][TESTS] Use `NoOp` datasource in SQL benchmarks ### What changes were proposed in this pull request? In the PR, I propose to replace `.collect()`, `.count()` and `.foreach(_ => ())` in SQL benchmarks and use the `NoOp` datasource. I added an implicit class to `SqlBasedBenchmark` with the `.noop()` method. It can be used in benchmark like: `ds.noop()`. The last one is unfolded to `ds.write.format("noop").mode(Overwrite).save()`. ### Why are the changes needed? To avoid additional overhead that `collect()` (and other actions) has. For example, `.collect()` has to convert values according to external types and pull data to the driver. This can hide actual performance regressions or improvements of benchmarked operations. ### Does this PR introduce any user-facing change? No ### How was this patch tested? Re-run all modified benchmarks using Amazon EC2. | Item | Description | | ---- | ----| | Region | us-west-2 (Oregon) | | Instance | r3.xlarge (spot instance) | | AMI | ami-06f2f779464715dc5 (ubuntu/images/hvm-ssd/ubuntu-bionic-18.04-amd64-server-20190722.1) | | Java | OpenJDK8/10 | - Run `TPCDSQueryBenchmark` using instructions from the PR #26049 ``` # `spark-tpcds-datagen` needs this. (JDK8) $ git clone https://github.com/apache/spark.git -b branch-2.4 --depth 1 spark-2.4 $ export SPARK_HOME=$PWD $ ./build/mvn clean package -DskipTests # Generate data. (JDK8) $ git clone gitgithub.com:maropu/spark-tpcds-datagen.git $ cd spark-tpcds-datagen/ $ build/mvn clean package $ mkdir -p /data/tpcds $ ./bin/dsdgen --output-location /data/tpcds/s1 // This need `Spark 2.4` ``` - Other benchmarks ran by the script: ``` #!/usr/bin/env python3 import os from sparktestsupport.shellutils import run_cmd benchmarks = [ ['sql/test', 'org.apache.spark.sql.execution.benchmark.AggregateBenchmark'], ['avro/test', 'org.apache.spark.sql.execution.benchmark.AvroReadBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.BloomFilterBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.DataSourceReadBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.DateTimeBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.ExtractBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.FilterPushdownBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.InExpressionBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.IntervalBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.JoinBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.MakeDateTimeBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.MiscBenchmark'], ['hive/test', 'org.apache.spark.sql.execution.benchmark.ObjectHashAggregateExecBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.OrcNestedSchemaPruningBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.OrcV2NestedSchemaPruningBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.ParquetNestedSchemaPruningBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.RangeBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.UDFBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.WideSchemaBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.WideTableBenchmark'], ['hive/test', 'org.apache.spark.sql.hive.orc.OrcReadBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.datasources.csv.CSVBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.datasources.json.JsonBenchmark'] ] print('Set SPARK_GENERATE_BENCHMARK_FILES=1') os.environ['SPARK_GENERATE_BENCHMARK_FILES'] = '1' for b in benchmarks: print("Run benchmark: %s" % b[1]) run_cmd(['build/sbt', '%s:runMain %s' % (b[0], b[1])]) ``` Closes #27078 from MaxGekk/noop-in-benchmarks. Lead-authored-by: Maxim Gekk <max.gekk@gmail.com> Co-authored-by: Maxim Gekk <maxim.gekk@databricks.com> Co-authored-by: Dongjoon Hyun <dhyun@apple.com> Signed-off-by: Dongjoon Hyun <dhyun@apple.com>
2020-01-12 16:18:19 -05:00
Data column - CSV 40273 40290 24 0.4 2560.5 1.0X
Data column - Json 14420 14440 28 1.1 916.8 2.8X
Data column - Parquet Vectorized 336 342 6 46.8 21.4 119.8X
Data column - Parquet MR 2651 2652 2 5.9 168.5 15.2X
Data column - ORC Vectorized 444 451 9 35.4 28.2 90.7X
Data column - ORC MR 2342 2356 20 6.7 148.9 17.2X
Partition column - CSV 11307 11310 4 1.4 718.9 3.6X
Partition column - Json 12105 12115 14 1.3 769.6 3.3X
Partition column - Parquet Vectorized 87 97 13 181.2 5.5 464.0X
Partition column - Parquet MR 1364 1368 7 11.5 86.7 29.5X
Partition column - ORC Vectorized 83 97 13 189.0 5.3 484.1X
Partition column - ORC MR 1424 1437 19 11.0 90.5 28.3X
Both columns - CSV 41896 42166 381 0.4 2663.7 1.0X
Both columns - Json 15852 15871 27 1.0 1007.8 2.5X
Both columns - Parquet Vectorized 379 383 5 41.5 24.1 106.2X
Both columns - Parquet MR 2889 2916 38 5.4 183.7 13.9X
Both columns - ORC Vectorized 581 582 2 27.1 36.9 69.3X
Both columns - ORC MR 2626 2641 22 6.0 166.9 15.3X
[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
================================================================================================
String with Nulls Scan
================================================================================================
[SPARK-30409][SPARK-29173][SQL][TESTS] Use `NoOp` datasource in SQL benchmarks ### What changes were proposed in this pull request? In the PR, I propose to replace `.collect()`, `.count()` and `.foreach(_ => ())` in SQL benchmarks and use the `NoOp` datasource. I added an implicit class to `SqlBasedBenchmark` with the `.noop()` method. It can be used in benchmark like: `ds.noop()`. The last one is unfolded to `ds.write.format("noop").mode(Overwrite).save()`. ### Why are the changes needed? To avoid additional overhead that `collect()` (and other actions) has. For example, `.collect()` has to convert values according to external types and pull data to the driver. This can hide actual performance regressions or improvements of benchmarked operations. ### Does this PR introduce any user-facing change? No ### How was this patch tested? Re-run all modified benchmarks using Amazon EC2. | Item | Description | | ---- | ----| | Region | us-west-2 (Oregon) | | Instance | r3.xlarge (spot instance) | | AMI | ami-06f2f779464715dc5 (ubuntu/images/hvm-ssd/ubuntu-bionic-18.04-amd64-server-20190722.1) | | Java | OpenJDK8/10 | - Run `TPCDSQueryBenchmark` using instructions from the PR #26049 ``` # `spark-tpcds-datagen` needs this. (JDK8) $ git clone https://github.com/apache/spark.git -b branch-2.4 --depth 1 spark-2.4 $ export SPARK_HOME=$PWD $ ./build/mvn clean package -DskipTests # Generate data. (JDK8) $ git clone gitgithub.com:maropu/spark-tpcds-datagen.git $ cd spark-tpcds-datagen/ $ build/mvn clean package $ mkdir -p /data/tpcds $ ./bin/dsdgen --output-location /data/tpcds/s1 // This need `Spark 2.4` ``` - Other benchmarks ran by the script: ``` #!/usr/bin/env python3 import os from sparktestsupport.shellutils import run_cmd benchmarks = [ ['sql/test', 'org.apache.spark.sql.execution.benchmark.AggregateBenchmark'], ['avro/test', 'org.apache.spark.sql.execution.benchmark.AvroReadBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.BloomFilterBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.DataSourceReadBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.DateTimeBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.ExtractBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.FilterPushdownBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.InExpressionBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.IntervalBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.JoinBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.MakeDateTimeBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.MiscBenchmark'], ['hive/test', 'org.apache.spark.sql.execution.benchmark.ObjectHashAggregateExecBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.OrcNestedSchemaPruningBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.OrcV2NestedSchemaPruningBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.ParquetNestedSchemaPruningBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.RangeBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.UDFBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.WideSchemaBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.WideTableBenchmark'], ['hive/test', 'org.apache.spark.sql.hive.orc.OrcReadBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.datasources.csv.CSVBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.datasources.json.JsonBenchmark'] ] print('Set SPARK_GENERATE_BENCHMARK_FILES=1') os.environ['SPARK_GENERATE_BENCHMARK_FILES'] = '1' for b in benchmarks: print("Run benchmark: %s" % b[1]) run_cmd(['build/sbt', '%s:runMain %s' % (b[0], b[1])]) ``` Closes #27078 from MaxGekk/noop-in-benchmarks. Lead-authored-by: Maxim Gekk <max.gekk@gmail.com> Co-authored-by: Maxim Gekk <maxim.gekk@databricks.com> Co-authored-by: Dongjoon Hyun <dhyun@apple.com> Signed-off-by: Dongjoon Hyun <dhyun@apple.com>
2020-01-12 16:18:19 -05:00
OpenJDK 64-Bit Server VM 11.0.5+10-post-Ubuntu-0ubuntu1.118.04 on Linux 4.15.0-1044-aws
[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
Intel(R) Xeon(R) CPU E5-2670 v2 @ 2.50GHz
String with Nulls Scan (0.0%): Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative
------------------------------------------------------------------------------------------------------------------------
[SPARK-30409][SPARK-29173][SQL][TESTS] Use `NoOp` datasource in SQL benchmarks ### What changes were proposed in this pull request? In the PR, I propose to replace `.collect()`, `.count()` and `.foreach(_ => ())` in SQL benchmarks and use the `NoOp` datasource. I added an implicit class to `SqlBasedBenchmark` with the `.noop()` method. It can be used in benchmark like: `ds.noop()`. The last one is unfolded to `ds.write.format("noop").mode(Overwrite).save()`. ### Why are the changes needed? To avoid additional overhead that `collect()` (and other actions) has. For example, `.collect()` has to convert values according to external types and pull data to the driver. This can hide actual performance regressions or improvements of benchmarked operations. ### Does this PR introduce any user-facing change? No ### How was this patch tested? Re-run all modified benchmarks using Amazon EC2. | Item | Description | | ---- | ----| | Region | us-west-2 (Oregon) | | Instance | r3.xlarge (spot instance) | | AMI | ami-06f2f779464715dc5 (ubuntu/images/hvm-ssd/ubuntu-bionic-18.04-amd64-server-20190722.1) | | Java | OpenJDK8/10 | - Run `TPCDSQueryBenchmark` using instructions from the PR #26049 ``` # `spark-tpcds-datagen` needs this. (JDK8) $ git clone https://github.com/apache/spark.git -b branch-2.4 --depth 1 spark-2.4 $ export SPARK_HOME=$PWD $ ./build/mvn clean package -DskipTests # Generate data. (JDK8) $ git clone gitgithub.com:maropu/spark-tpcds-datagen.git $ cd spark-tpcds-datagen/ $ build/mvn clean package $ mkdir -p /data/tpcds $ ./bin/dsdgen --output-location /data/tpcds/s1 // This need `Spark 2.4` ``` - Other benchmarks ran by the script: ``` #!/usr/bin/env python3 import os from sparktestsupport.shellutils import run_cmd benchmarks = [ ['sql/test', 'org.apache.spark.sql.execution.benchmark.AggregateBenchmark'], ['avro/test', 'org.apache.spark.sql.execution.benchmark.AvroReadBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.BloomFilterBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.DataSourceReadBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.DateTimeBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.ExtractBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.FilterPushdownBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.InExpressionBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.IntervalBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.JoinBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.MakeDateTimeBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.MiscBenchmark'], ['hive/test', 'org.apache.spark.sql.execution.benchmark.ObjectHashAggregateExecBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.OrcNestedSchemaPruningBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.OrcV2NestedSchemaPruningBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.ParquetNestedSchemaPruningBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.RangeBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.UDFBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.WideSchemaBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.WideTableBenchmark'], ['hive/test', 'org.apache.spark.sql.hive.orc.OrcReadBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.datasources.csv.CSVBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.datasources.json.JsonBenchmark'] ] print('Set SPARK_GENERATE_BENCHMARK_FILES=1') os.environ['SPARK_GENERATE_BENCHMARK_FILES'] = '1' for b in benchmarks: print("Run benchmark: %s" % b[1]) run_cmd(['build/sbt', '%s:runMain %s' % (b[0], b[1])]) ``` Closes #27078 from MaxGekk/noop-in-benchmarks. Lead-authored-by: Maxim Gekk <max.gekk@gmail.com> Co-authored-by: Maxim Gekk <maxim.gekk@databricks.com> Co-authored-by: Dongjoon Hyun <dhyun@apple.com> Signed-off-by: Dongjoon Hyun <dhyun@apple.com>
2020-01-12 16:18:19 -05:00
SQL CSV 20831 21141 439 0.5 1986.6 1.0X
SQL Json 11720 11721 1 0.9 1117.7 1.8X
SQL Parquet Vectorized 1470 1475 7 7.1 140.2 14.2X
SQL Parquet MR 3902 3902 0 2.7 372.1 5.3X
ParquetReader Vectorized 1074 1077 4 9.8 102.5 19.4X
SQL ORC Vectorized 1289 1334 64 8.1 122.9 16.2X
SQL ORC MR 3603 3612 13 2.9 343.6 5.8X
[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-30409][SPARK-29173][SQL][TESTS] Use `NoOp` datasource in SQL benchmarks ### What changes were proposed in this pull request? In the PR, I propose to replace `.collect()`, `.count()` and `.foreach(_ => ())` in SQL benchmarks and use the `NoOp` datasource. I added an implicit class to `SqlBasedBenchmark` with the `.noop()` method. It can be used in benchmark like: `ds.noop()`. The last one is unfolded to `ds.write.format("noop").mode(Overwrite).save()`. ### Why are the changes needed? To avoid additional overhead that `collect()` (and other actions) has. For example, `.collect()` has to convert values according to external types and pull data to the driver. This can hide actual performance regressions or improvements of benchmarked operations. ### Does this PR introduce any user-facing change? No ### How was this patch tested? Re-run all modified benchmarks using Amazon EC2. | Item | Description | | ---- | ----| | Region | us-west-2 (Oregon) | | Instance | r3.xlarge (spot instance) | | AMI | ami-06f2f779464715dc5 (ubuntu/images/hvm-ssd/ubuntu-bionic-18.04-amd64-server-20190722.1) | | Java | OpenJDK8/10 | - Run `TPCDSQueryBenchmark` using instructions from the PR #26049 ``` # `spark-tpcds-datagen` needs this. (JDK8) $ git clone https://github.com/apache/spark.git -b branch-2.4 --depth 1 spark-2.4 $ export SPARK_HOME=$PWD $ ./build/mvn clean package -DskipTests # Generate data. (JDK8) $ git clone gitgithub.com:maropu/spark-tpcds-datagen.git $ cd spark-tpcds-datagen/ $ build/mvn clean package $ mkdir -p /data/tpcds $ ./bin/dsdgen --output-location /data/tpcds/s1 // This need `Spark 2.4` ``` - Other benchmarks ran by the script: ``` #!/usr/bin/env python3 import os from sparktestsupport.shellutils import run_cmd benchmarks = [ ['sql/test', 'org.apache.spark.sql.execution.benchmark.AggregateBenchmark'], ['avro/test', 'org.apache.spark.sql.execution.benchmark.AvroReadBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.BloomFilterBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.DataSourceReadBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.DateTimeBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.ExtractBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.FilterPushdownBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.InExpressionBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.IntervalBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.JoinBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.MakeDateTimeBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.MiscBenchmark'], ['hive/test', 'org.apache.spark.sql.execution.benchmark.ObjectHashAggregateExecBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.OrcNestedSchemaPruningBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.OrcV2NestedSchemaPruningBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.ParquetNestedSchemaPruningBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.RangeBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.UDFBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.WideSchemaBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.WideTableBenchmark'], ['hive/test', 'org.apache.spark.sql.hive.orc.OrcReadBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.datasources.csv.CSVBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.datasources.json.JsonBenchmark'] ] print('Set SPARK_GENERATE_BENCHMARK_FILES=1') os.environ['SPARK_GENERATE_BENCHMARK_FILES'] = '1' for b in benchmarks: print("Run benchmark: %s" % b[1]) run_cmd(['build/sbt', '%s:runMain %s' % (b[0], b[1])]) ``` Closes #27078 from MaxGekk/noop-in-benchmarks. Lead-authored-by: Maxim Gekk <max.gekk@gmail.com> Co-authored-by: Maxim Gekk <maxim.gekk@databricks.com> Co-authored-by: Dongjoon Hyun <dhyun@apple.com> Signed-off-by: Dongjoon Hyun <dhyun@apple.com>
2020-01-12 16:18:19 -05:00
OpenJDK 64-Bit Server VM 11.0.5+10-post-Ubuntu-0ubuntu1.118.04 on Linux 4.15.0-1044-aws
[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
Intel(R) Xeon(R) CPU E5-2670 v2 @ 2.50GHz
String with Nulls Scan (50.0%): Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative
------------------------------------------------------------------------------------------------------------------------
[SPARK-30409][SPARK-29173][SQL][TESTS] Use `NoOp` datasource in SQL benchmarks ### What changes were proposed in this pull request? In the PR, I propose to replace `.collect()`, `.count()` and `.foreach(_ => ())` in SQL benchmarks and use the `NoOp` datasource. I added an implicit class to `SqlBasedBenchmark` with the `.noop()` method. It can be used in benchmark like: `ds.noop()`. The last one is unfolded to `ds.write.format("noop").mode(Overwrite).save()`. ### Why are the changes needed? To avoid additional overhead that `collect()` (and other actions) has. For example, `.collect()` has to convert values according to external types and pull data to the driver. This can hide actual performance regressions or improvements of benchmarked operations. ### Does this PR introduce any user-facing change? No ### How was this patch tested? Re-run all modified benchmarks using Amazon EC2. | Item | Description | | ---- | ----| | Region | us-west-2 (Oregon) | | Instance | r3.xlarge (spot instance) | | AMI | ami-06f2f779464715dc5 (ubuntu/images/hvm-ssd/ubuntu-bionic-18.04-amd64-server-20190722.1) | | Java | OpenJDK8/10 | - Run `TPCDSQueryBenchmark` using instructions from the PR #26049 ``` # `spark-tpcds-datagen` needs this. (JDK8) $ git clone https://github.com/apache/spark.git -b branch-2.4 --depth 1 spark-2.4 $ export SPARK_HOME=$PWD $ ./build/mvn clean package -DskipTests # Generate data. (JDK8) $ git clone gitgithub.com:maropu/spark-tpcds-datagen.git $ cd spark-tpcds-datagen/ $ build/mvn clean package $ mkdir -p /data/tpcds $ ./bin/dsdgen --output-location /data/tpcds/s1 // This need `Spark 2.4` ``` - Other benchmarks ran by the script: ``` #!/usr/bin/env python3 import os from sparktestsupport.shellutils import run_cmd benchmarks = [ ['sql/test', 'org.apache.spark.sql.execution.benchmark.AggregateBenchmark'], ['avro/test', 'org.apache.spark.sql.execution.benchmark.AvroReadBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.BloomFilterBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.DataSourceReadBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.DateTimeBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.ExtractBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.FilterPushdownBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.InExpressionBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.IntervalBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.JoinBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.MakeDateTimeBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.MiscBenchmark'], ['hive/test', 'org.apache.spark.sql.execution.benchmark.ObjectHashAggregateExecBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.OrcNestedSchemaPruningBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.OrcV2NestedSchemaPruningBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.ParquetNestedSchemaPruningBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.RangeBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.UDFBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.WideSchemaBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.WideTableBenchmark'], ['hive/test', 'org.apache.spark.sql.hive.orc.OrcReadBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.datasources.csv.CSVBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.datasources.json.JsonBenchmark'] ] print('Set SPARK_GENERATE_BENCHMARK_FILES=1') os.environ['SPARK_GENERATE_BENCHMARK_FILES'] = '1' for b in benchmarks: print("Run benchmark: %s" % b[1]) run_cmd(['build/sbt', '%s:runMain %s' % (b[0], b[1])]) ``` Closes #27078 from MaxGekk/noop-in-benchmarks. Lead-authored-by: Maxim Gekk <max.gekk@gmail.com> Co-authored-by: Maxim Gekk <maxim.gekk@databricks.com> Co-authored-by: Dongjoon Hyun <dhyun@apple.com> Signed-off-by: Dongjoon Hyun <dhyun@apple.com>
2020-01-12 16:18:19 -05:00
SQL CSV 21850 21910 85 0.5 2083.8 1.0X
SQL Json 8651 8668 24 1.2 825.0 2.5X
SQL Parquet Vectorized 1079 1090 16 9.7 102.9 20.3X
SQL Parquet MR 2906 2925 27 3.6 277.1 7.5X
ParquetReader Vectorized 951 954 4 11.0 90.7 23.0X
SQL ORC Vectorized 1246 1250 5 8.4 118.8 17.5X
SQL ORC MR 3146 3162 22 3.3 300.1 6.9X
[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-30409][SPARK-29173][SQL][TESTS] Use `NoOp` datasource in SQL benchmarks ### What changes were proposed in this pull request? In the PR, I propose to replace `.collect()`, `.count()` and `.foreach(_ => ())` in SQL benchmarks and use the `NoOp` datasource. I added an implicit class to `SqlBasedBenchmark` with the `.noop()` method. It can be used in benchmark like: `ds.noop()`. The last one is unfolded to `ds.write.format("noop").mode(Overwrite).save()`. ### Why are the changes needed? To avoid additional overhead that `collect()` (and other actions) has. For example, `.collect()` has to convert values according to external types and pull data to the driver. This can hide actual performance regressions or improvements of benchmarked operations. ### Does this PR introduce any user-facing change? No ### How was this patch tested? Re-run all modified benchmarks using Amazon EC2. | Item | Description | | ---- | ----| | Region | us-west-2 (Oregon) | | Instance | r3.xlarge (spot instance) | | AMI | ami-06f2f779464715dc5 (ubuntu/images/hvm-ssd/ubuntu-bionic-18.04-amd64-server-20190722.1) | | Java | OpenJDK8/10 | - Run `TPCDSQueryBenchmark` using instructions from the PR #26049 ``` # `spark-tpcds-datagen` needs this. (JDK8) $ git clone https://github.com/apache/spark.git -b branch-2.4 --depth 1 spark-2.4 $ export SPARK_HOME=$PWD $ ./build/mvn clean package -DskipTests # Generate data. (JDK8) $ git clone gitgithub.com:maropu/spark-tpcds-datagen.git $ cd spark-tpcds-datagen/ $ build/mvn clean package $ mkdir -p /data/tpcds $ ./bin/dsdgen --output-location /data/tpcds/s1 // This need `Spark 2.4` ``` - Other benchmarks ran by the script: ``` #!/usr/bin/env python3 import os from sparktestsupport.shellutils import run_cmd benchmarks = [ ['sql/test', 'org.apache.spark.sql.execution.benchmark.AggregateBenchmark'], ['avro/test', 'org.apache.spark.sql.execution.benchmark.AvroReadBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.BloomFilterBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.DataSourceReadBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.DateTimeBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.ExtractBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.FilterPushdownBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.InExpressionBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.IntervalBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.JoinBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.MakeDateTimeBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.MiscBenchmark'], ['hive/test', 'org.apache.spark.sql.execution.benchmark.ObjectHashAggregateExecBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.OrcNestedSchemaPruningBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.OrcV2NestedSchemaPruningBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.ParquetNestedSchemaPruningBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.RangeBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.UDFBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.WideSchemaBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.WideTableBenchmark'], ['hive/test', 'org.apache.spark.sql.hive.orc.OrcReadBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.datasources.csv.CSVBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.datasources.json.JsonBenchmark'] ] print('Set SPARK_GENERATE_BENCHMARK_FILES=1') os.environ['SPARK_GENERATE_BENCHMARK_FILES'] = '1' for b in benchmarks: print("Run benchmark: %s" % b[1]) run_cmd(['build/sbt', '%s:runMain %s' % (b[0], b[1])]) ``` Closes #27078 from MaxGekk/noop-in-benchmarks. Lead-authored-by: Maxim Gekk <max.gekk@gmail.com> Co-authored-by: Maxim Gekk <maxim.gekk@databricks.com> Co-authored-by: Dongjoon Hyun <dhyun@apple.com> Signed-off-by: Dongjoon Hyun <dhyun@apple.com>
2020-01-12 16:18:19 -05:00
OpenJDK 64-Bit Server VM 11.0.5+10-post-Ubuntu-0ubuntu1.118.04 on Linux 4.15.0-1044-aws
[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
Intel(R) Xeon(R) CPU E5-2670 v2 @ 2.50GHz
String with Nulls Scan (95.0%): Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative
------------------------------------------------------------------------------------------------------------------------
[SPARK-30409][SPARK-29173][SQL][TESTS] Use `NoOp` datasource in SQL benchmarks ### What changes were proposed in this pull request? In the PR, I propose to replace `.collect()`, `.count()` and `.foreach(_ => ())` in SQL benchmarks and use the `NoOp` datasource. I added an implicit class to `SqlBasedBenchmark` with the `.noop()` method. It can be used in benchmark like: `ds.noop()`. The last one is unfolded to `ds.write.format("noop").mode(Overwrite).save()`. ### Why are the changes needed? To avoid additional overhead that `collect()` (and other actions) has. For example, `.collect()` has to convert values according to external types and pull data to the driver. This can hide actual performance regressions or improvements of benchmarked operations. ### Does this PR introduce any user-facing change? No ### How was this patch tested? Re-run all modified benchmarks using Amazon EC2. | Item | Description | | ---- | ----| | Region | us-west-2 (Oregon) | | Instance | r3.xlarge (spot instance) | | AMI | ami-06f2f779464715dc5 (ubuntu/images/hvm-ssd/ubuntu-bionic-18.04-amd64-server-20190722.1) | | Java | OpenJDK8/10 | - Run `TPCDSQueryBenchmark` using instructions from the PR #26049 ``` # `spark-tpcds-datagen` needs this. (JDK8) $ git clone https://github.com/apache/spark.git -b branch-2.4 --depth 1 spark-2.4 $ export SPARK_HOME=$PWD $ ./build/mvn clean package -DskipTests # Generate data. (JDK8) $ git clone gitgithub.com:maropu/spark-tpcds-datagen.git $ cd spark-tpcds-datagen/ $ build/mvn clean package $ mkdir -p /data/tpcds $ ./bin/dsdgen --output-location /data/tpcds/s1 // This need `Spark 2.4` ``` - Other benchmarks ran by the script: ``` #!/usr/bin/env python3 import os from sparktestsupport.shellutils import run_cmd benchmarks = [ ['sql/test', 'org.apache.spark.sql.execution.benchmark.AggregateBenchmark'], ['avro/test', 'org.apache.spark.sql.execution.benchmark.AvroReadBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.BloomFilterBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.DataSourceReadBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.DateTimeBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.ExtractBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.FilterPushdownBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.InExpressionBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.IntervalBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.JoinBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.MakeDateTimeBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.MiscBenchmark'], ['hive/test', 'org.apache.spark.sql.execution.benchmark.ObjectHashAggregateExecBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.OrcNestedSchemaPruningBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.OrcV2NestedSchemaPruningBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.ParquetNestedSchemaPruningBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.RangeBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.UDFBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.WideSchemaBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.WideTableBenchmark'], ['hive/test', 'org.apache.spark.sql.hive.orc.OrcReadBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.datasources.csv.CSVBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.datasources.json.JsonBenchmark'] ] print('Set SPARK_GENERATE_BENCHMARK_FILES=1') os.environ['SPARK_GENERATE_BENCHMARK_FILES'] = '1' for b in benchmarks: print("Run benchmark: %s" % b[1]) run_cmd(['build/sbt', '%s:runMain %s' % (b[0], b[1])]) ``` Closes #27078 from MaxGekk/noop-in-benchmarks. Lead-authored-by: Maxim Gekk <max.gekk@gmail.com> Co-authored-by: Maxim Gekk <maxim.gekk@databricks.com> Co-authored-by: Dongjoon Hyun <dhyun@apple.com> Signed-off-by: Dongjoon Hyun <dhyun@apple.com>
2020-01-12 16:18:19 -05:00
SQL CSV 18993 19140 209 0.6 1811.3 1.0X
SQL Json 5467 5469 2 1.9 521.4 3.5X
SQL Parquet Vectorized 240 248 10 43.8 22.8 79.3X
SQL Parquet MR 1745 1753 12 6.0 166.4 10.9X
ParquetReader Vectorized 240 244 5 43.7 22.9 79.1X
SQL ORC Vectorized 496 500 4 21.1 47.3 38.3X
SQL ORC MR 1822 1827 8 5.8 173.7 10.4X
[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
================================================================================================
Single Column Scan From Wide Columns
================================================================================================
[SPARK-30409][SPARK-29173][SQL][TESTS] Use `NoOp` datasource in SQL benchmarks ### What changes were proposed in this pull request? In the PR, I propose to replace `.collect()`, `.count()` and `.foreach(_ => ())` in SQL benchmarks and use the `NoOp` datasource. I added an implicit class to `SqlBasedBenchmark` with the `.noop()` method. It can be used in benchmark like: `ds.noop()`. The last one is unfolded to `ds.write.format("noop").mode(Overwrite).save()`. ### Why are the changes needed? To avoid additional overhead that `collect()` (and other actions) has. For example, `.collect()` has to convert values according to external types and pull data to the driver. This can hide actual performance regressions or improvements of benchmarked operations. ### Does this PR introduce any user-facing change? No ### How was this patch tested? Re-run all modified benchmarks using Amazon EC2. | Item | Description | | ---- | ----| | Region | us-west-2 (Oregon) | | Instance | r3.xlarge (spot instance) | | AMI | ami-06f2f779464715dc5 (ubuntu/images/hvm-ssd/ubuntu-bionic-18.04-amd64-server-20190722.1) | | Java | OpenJDK8/10 | - Run `TPCDSQueryBenchmark` using instructions from the PR #26049 ``` # `spark-tpcds-datagen` needs this. (JDK8) $ git clone https://github.com/apache/spark.git -b branch-2.4 --depth 1 spark-2.4 $ export SPARK_HOME=$PWD $ ./build/mvn clean package -DskipTests # Generate data. (JDK8) $ git clone gitgithub.com:maropu/spark-tpcds-datagen.git $ cd spark-tpcds-datagen/ $ build/mvn clean package $ mkdir -p /data/tpcds $ ./bin/dsdgen --output-location /data/tpcds/s1 // This need `Spark 2.4` ``` - Other benchmarks ran by the script: ``` #!/usr/bin/env python3 import os from sparktestsupport.shellutils import run_cmd benchmarks = [ ['sql/test', 'org.apache.spark.sql.execution.benchmark.AggregateBenchmark'], ['avro/test', 'org.apache.spark.sql.execution.benchmark.AvroReadBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.BloomFilterBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.DataSourceReadBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.DateTimeBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.ExtractBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.FilterPushdownBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.InExpressionBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.IntervalBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.JoinBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.MakeDateTimeBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.MiscBenchmark'], ['hive/test', 'org.apache.spark.sql.execution.benchmark.ObjectHashAggregateExecBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.OrcNestedSchemaPruningBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.OrcV2NestedSchemaPruningBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.ParquetNestedSchemaPruningBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.RangeBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.UDFBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.WideSchemaBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.WideTableBenchmark'], ['hive/test', 'org.apache.spark.sql.hive.orc.OrcReadBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.datasources.csv.CSVBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.datasources.json.JsonBenchmark'] ] print('Set SPARK_GENERATE_BENCHMARK_FILES=1') os.environ['SPARK_GENERATE_BENCHMARK_FILES'] = '1' for b in benchmarks: print("Run benchmark: %s" % b[1]) run_cmd(['build/sbt', '%s:runMain %s' % (b[0], b[1])]) ``` Closes #27078 from MaxGekk/noop-in-benchmarks. Lead-authored-by: Maxim Gekk <max.gekk@gmail.com> Co-authored-by: Maxim Gekk <maxim.gekk@databricks.com> Co-authored-by: Dongjoon Hyun <dhyun@apple.com> Signed-off-by: Dongjoon Hyun <dhyun@apple.com>
2020-01-12 16:18:19 -05:00
OpenJDK 64-Bit Server VM 11.0.5+10-post-Ubuntu-0ubuntu1.118.04 on Linux 4.15.0-1044-aws
[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
Intel(R) Xeon(R) CPU E5-2670 v2 @ 2.50GHz
Single Column Scan from 10 columns: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative
------------------------------------------------------------------------------------------------------------------------
[SPARK-30409][SPARK-29173][SQL][TESTS] Use `NoOp` datasource in SQL benchmarks ### What changes were proposed in this pull request? In the PR, I propose to replace `.collect()`, `.count()` and `.foreach(_ => ())` in SQL benchmarks and use the `NoOp` datasource. I added an implicit class to `SqlBasedBenchmark` with the `.noop()` method. It can be used in benchmark like: `ds.noop()`. The last one is unfolded to `ds.write.format("noop").mode(Overwrite).save()`. ### Why are the changes needed? To avoid additional overhead that `collect()` (and other actions) has. For example, `.collect()` has to convert values according to external types and pull data to the driver. This can hide actual performance regressions or improvements of benchmarked operations. ### Does this PR introduce any user-facing change? No ### How was this patch tested? Re-run all modified benchmarks using Amazon EC2. | Item | Description | | ---- | ----| | Region | us-west-2 (Oregon) | | Instance | r3.xlarge (spot instance) | | AMI | ami-06f2f779464715dc5 (ubuntu/images/hvm-ssd/ubuntu-bionic-18.04-amd64-server-20190722.1) | | Java | OpenJDK8/10 | - Run `TPCDSQueryBenchmark` using instructions from the PR #26049 ``` # `spark-tpcds-datagen` needs this. (JDK8) $ git clone https://github.com/apache/spark.git -b branch-2.4 --depth 1 spark-2.4 $ export SPARK_HOME=$PWD $ ./build/mvn clean package -DskipTests # Generate data. (JDK8) $ git clone gitgithub.com:maropu/spark-tpcds-datagen.git $ cd spark-tpcds-datagen/ $ build/mvn clean package $ mkdir -p /data/tpcds $ ./bin/dsdgen --output-location /data/tpcds/s1 // This need `Spark 2.4` ``` - Other benchmarks ran by the script: ``` #!/usr/bin/env python3 import os from sparktestsupport.shellutils import run_cmd benchmarks = [ ['sql/test', 'org.apache.spark.sql.execution.benchmark.AggregateBenchmark'], ['avro/test', 'org.apache.spark.sql.execution.benchmark.AvroReadBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.BloomFilterBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.DataSourceReadBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.DateTimeBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.ExtractBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.FilterPushdownBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.InExpressionBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.IntervalBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.JoinBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.MakeDateTimeBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.MiscBenchmark'], ['hive/test', 'org.apache.spark.sql.execution.benchmark.ObjectHashAggregateExecBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.OrcNestedSchemaPruningBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.OrcV2NestedSchemaPruningBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.ParquetNestedSchemaPruningBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.RangeBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.UDFBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.WideSchemaBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.WideTableBenchmark'], ['hive/test', 'org.apache.spark.sql.hive.orc.OrcReadBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.datasources.csv.CSVBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.datasources.json.JsonBenchmark'] ] print('Set SPARK_GENERATE_BENCHMARK_FILES=1') os.environ['SPARK_GENERATE_BENCHMARK_FILES'] = '1' for b in benchmarks: print("Run benchmark: %s" % b[1]) run_cmd(['build/sbt', '%s:runMain %s' % (b[0], b[1])]) ``` Closes #27078 from MaxGekk/noop-in-benchmarks. Lead-authored-by: Maxim Gekk <max.gekk@gmail.com> Co-authored-by: Maxim Gekk <maxim.gekk@databricks.com> Co-authored-by: Dongjoon Hyun <dhyun@apple.com> Signed-off-by: Dongjoon Hyun <dhyun@apple.com>
2020-01-12 16:18:19 -05:00
SQL CSV 3907 3911 6 0.3 3726.3 1.0X
SQL Json 3755 3763 12 0.3 3581.2 1.0X
SQL Parquet Vectorized 68 71 6 15.4 64.8 57.5X
SQL Parquet MR 234 239 5 4.5 223.0 16.7X
SQL ORC Vectorized 74 77 5 14.2 70.4 52.9X
SQL ORC MR 203 204 2 5.2 193.3 19.3X
[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-30409][SPARK-29173][SQL][TESTS] Use `NoOp` datasource in SQL benchmarks ### What changes were proposed in this pull request? In the PR, I propose to replace `.collect()`, `.count()` and `.foreach(_ => ())` in SQL benchmarks and use the `NoOp` datasource. I added an implicit class to `SqlBasedBenchmark` with the `.noop()` method. It can be used in benchmark like: `ds.noop()`. The last one is unfolded to `ds.write.format("noop").mode(Overwrite).save()`. ### Why are the changes needed? To avoid additional overhead that `collect()` (and other actions) has. For example, `.collect()` has to convert values according to external types and pull data to the driver. This can hide actual performance regressions or improvements of benchmarked operations. ### Does this PR introduce any user-facing change? No ### How was this patch tested? Re-run all modified benchmarks using Amazon EC2. | Item | Description | | ---- | ----| | Region | us-west-2 (Oregon) | | Instance | r3.xlarge (spot instance) | | AMI | ami-06f2f779464715dc5 (ubuntu/images/hvm-ssd/ubuntu-bionic-18.04-amd64-server-20190722.1) | | Java | OpenJDK8/10 | - Run `TPCDSQueryBenchmark` using instructions from the PR #26049 ``` # `spark-tpcds-datagen` needs this. (JDK8) $ git clone https://github.com/apache/spark.git -b branch-2.4 --depth 1 spark-2.4 $ export SPARK_HOME=$PWD $ ./build/mvn clean package -DskipTests # Generate data. (JDK8) $ git clone gitgithub.com:maropu/spark-tpcds-datagen.git $ cd spark-tpcds-datagen/ $ build/mvn clean package $ mkdir -p /data/tpcds $ ./bin/dsdgen --output-location /data/tpcds/s1 // This need `Spark 2.4` ``` - Other benchmarks ran by the script: ``` #!/usr/bin/env python3 import os from sparktestsupport.shellutils import run_cmd benchmarks = [ ['sql/test', 'org.apache.spark.sql.execution.benchmark.AggregateBenchmark'], ['avro/test', 'org.apache.spark.sql.execution.benchmark.AvroReadBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.BloomFilterBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.DataSourceReadBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.DateTimeBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.ExtractBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.FilterPushdownBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.InExpressionBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.IntervalBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.JoinBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.MakeDateTimeBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.MiscBenchmark'], ['hive/test', 'org.apache.spark.sql.execution.benchmark.ObjectHashAggregateExecBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.OrcNestedSchemaPruningBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.OrcV2NestedSchemaPruningBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.ParquetNestedSchemaPruningBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.RangeBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.UDFBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.WideSchemaBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.WideTableBenchmark'], ['hive/test', 'org.apache.spark.sql.hive.orc.OrcReadBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.datasources.csv.CSVBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.datasources.json.JsonBenchmark'] ] print('Set SPARK_GENERATE_BENCHMARK_FILES=1') os.environ['SPARK_GENERATE_BENCHMARK_FILES'] = '1' for b in benchmarks: print("Run benchmark: %s" % b[1]) run_cmd(['build/sbt', '%s:runMain %s' % (b[0], b[1])]) ``` Closes #27078 from MaxGekk/noop-in-benchmarks. Lead-authored-by: Maxim Gekk <max.gekk@gmail.com> Co-authored-by: Maxim Gekk <maxim.gekk@databricks.com> Co-authored-by: Dongjoon Hyun <dhyun@apple.com> Signed-off-by: Dongjoon Hyun <dhyun@apple.com>
2020-01-12 16:18:19 -05:00
OpenJDK 64-Bit Server VM 11.0.5+10-post-Ubuntu-0ubuntu1.118.04 on Linux 4.15.0-1044-aws
[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
Intel(R) Xeon(R) CPU E5-2670 v2 @ 2.50GHz
Single Column Scan from 50 columns: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative
------------------------------------------------------------------------------------------------------------------------
[SPARK-30409][SPARK-29173][SQL][TESTS] Use `NoOp` datasource in SQL benchmarks ### What changes were proposed in this pull request? In the PR, I propose to replace `.collect()`, `.count()` and `.foreach(_ => ())` in SQL benchmarks and use the `NoOp` datasource. I added an implicit class to `SqlBasedBenchmark` with the `.noop()` method. It can be used in benchmark like: `ds.noop()`. The last one is unfolded to `ds.write.format("noop").mode(Overwrite).save()`. ### Why are the changes needed? To avoid additional overhead that `collect()` (and other actions) has. For example, `.collect()` has to convert values according to external types and pull data to the driver. This can hide actual performance regressions or improvements of benchmarked operations. ### Does this PR introduce any user-facing change? No ### How was this patch tested? Re-run all modified benchmarks using Amazon EC2. | Item | Description | | ---- | ----| | Region | us-west-2 (Oregon) | | Instance | r3.xlarge (spot instance) | | AMI | ami-06f2f779464715dc5 (ubuntu/images/hvm-ssd/ubuntu-bionic-18.04-amd64-server-20190722.1) | | Java | OpenJDK8/10 | - Run `TPCDSQueryBenchmark` using instructions from the PR #26049 ``` # `spark-tpcds-datagen` needs this. (JDK8) $ git clone https://github.com/apache/spark.git -b branch-2.4 --depth 1 spark-2.4 $ export SPARK_HOME=$PWD $ ./build/mvn clean package -DskipTests # Generate data. (JDK8) $ git clone gitgithub.com:maropu/spark-tpcds-datagen.git $ cd spark-tpcds-datagen/ $ build/mvn clean package $ mkdir -p /data/tpcds $ ./bin/dsdgen --output-location /data/tpcds/s1 // This need `Spark 2.4` ``` - Other benchmarks ran by the script: ``` #!/usr/bin/env python3 import os from sparktestsupport.shellutils import run_cmd benchmarks = [ ['sql/test', 'org.apache.spark.sql.execution.benchmark.AggregateBenchmark'], ['avro/test', 'org.apache.spark.sql.execution.benchmark.AvroReadBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.BloomFilterBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.DataSourceReadBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.DateTimeBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.ExtractBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.FilterPushdownBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.InExpressionBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.IntervalBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.JoinBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.MakeDateTimeBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.MiscBenchmark'], ['hive/test', 'org.apache.spark.sql.execution.benchmark.ObjectHashAggregateExecBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.OrcNestedSchemaPruningBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.OrcV2NestedSchemaPruningBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.ParquetNestedSchemaPruningBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.RangeBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.UDFBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.WideSchemaBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.WideTableBenchmark'], ['hive/test', 'org.apache.spark.sql.hive.orc.OrcReadBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.datasources.csv.CSVBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.datasources.json.JsonBenchmark'] ] print('Set SPARK_GENERATE_BENCHMARK_FILES=1') os.environ['SPARK_GENERATE_BENCHMARK_FILES'] = '1' for b in benchmarks: print("Run benchmark: %s" % b[1]) run_cmd(['build/sbt', '%s:runMain %s' % (b[0], b[1])]) ``` Closes #27078 from MaxGekk/noop-in-benchmarks. Lead-authored-by: Maxim Gekk <max.gekk@gmail.com> Co-authored-by: Maxim Gekk <maxim.gekk@databricks.com> Co-authored-by: Dongjoon Hyun <dhyun@apple.com> Signed-off-by: Dongjoon Hyun <dhyun@apple.com>
2020-01-12 16:18:19 -05:00
SQL CSV 7909 7927 25 0.1 7542.9 1.0X
SQL Json 15014 15101 123 0.1 14318.8 0.5X
SQL Parquet Vectorized 105 128 22 10.0 100.0 75.4X
SQL Parquet MR 275 283 9 3.8 261.9 28.8X
SQL ORC Vectorized 104 116 9 10.1 98.9 76.3X
SQL ORC MR 234 245 12 4.5 223.0 33.8X
[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-30409][SPARK-29173][SQL][TESTS] Use `NoOp` datasource in SQL benchmarks ### What changes were proposed in this pull request? In the PR, I propose to replace `.collect()`, `.count()` and `.foreach(_ => ())` in SQL benchmarks and use the `NoOp` datasource. I added an implicit class to `SqlBasedBenchmark` with the `.noop()` method. It can be used in benchmark like: `ds.noop()`. The last one is unfolded to `ds.write.format("noop").mode(Overwrite).save()`. ### Why are the changes needed? To avoid additional overhead that `collect()` (and other actions) has. For example, `.collect()` has to convert values according to external types and pull data to the driver. This can hide actual performance regressions or improvements of benchmarked operations. ### Does this PR introduce any user-facing change? No ### How was this patch tested? Re-run all modified benchmarks using Amazon EC2. | Item | Description | | ---- | ----| | Region | us-west-2 (Oregon) | | Instance | r3.xlarge (spot instance) | | AMI | ami-06f2f779464715dc5 (ubuntu/images/hvm-ssd/ubuntu-bionic-18.04-amd64-server-20190722.1) | | Java | OpenJDK8/10 | - Run `TPCDSQueryBenchmark` using instructions from the PR #26049 ``` # `spark-tpcds-datagen` needs this. (JDK8) $ git clone https://github.com/apache/spark.git -b branch-2.4 --depth 1 spark-2.4 $ export SPARK_HOME=$PWD $ ./build/mvn clean package -DskipTests # Generate data. (JDK8) $ git clone gitgithub.com:maropu/spark-tpcds-datagen.git $ cd spark-tpcds-datagen/ $ build/mvn clean package $ mkdir -p /data/tpcds $ ./bin/dsdgen --output-location /data/tpcds/s1 // This need `Spark 2.4` ``` - Other benchmarks ran by the script: ``` #!/usr/bin/env python3 import os from sparktestsupport.shellutils import run_cmd benchmarks = [ ['sql/test', 'org.apache.spark.sql.execution.benchmark.AggregateBenchmark'], ['avro/test', 'org.apache.spark.sql.execution.benchmark.AvroReadBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.BloomFilterBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.DataSourceReadBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.DateTimeBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.ExtractBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.FilterPushdownBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.InExpressionBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.IntervalBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.JoinBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.MakeDateTimeBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.MiscBenchmark'], ['hive/test', 'org.apache.spark.sql.execution.benchmark.ObjectHashAggregateExecBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.OrcNestedSchemaPruningBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.OrcV2NestedSchemaPruningBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.ParquetNestedSchemaPruningBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.RangeBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.UDFBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.WideSchemaBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.WideTableBenchmark'], ['hive/test', 'org.apache.spark.sql.hive.orc.OrcReadBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.datasources.csv.CSVBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.datasources.json.JsonBenchmark'] ] print('Set SPARK_GENERATE_BENCHMARK_FILES=1') os.environ['SPARK_GENERATE_BENCHMARK_FILES'] = '1' for b in benchmarks: print("Run benchmark: %s" % b[1]) run_cmd(['build/sbt', '%s:runMain %s' % (b[0], b[1])]) ``` Closes #27078 from MaxGekk/noop-in-benchmarks. Lead-authored-by: Maxim Gekk <max.gekk@gmail.com> Co-authored-by: Maxim Gekk <maxim.gekk@databricks.com> Co-authored-by: Dongjoon Hyun <dhyun@apple.com> Signed-off-by: Dongjoon Hyun <dhyun@apple.com>
2020-01-12 16:18:19 -05:00
OpenJDK 64-Bit Server VM 11.0.5+10-post-Ubuntu-0ubuntu1.118.04 on Linux 4.15.0-1044-aws
[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
Intel(R) Xeon(R) CPU E5-2670 v2 @ 2.50GHz
Single Column Scan from 100 columns: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative
------------------------------------------------------------------------------------------------------------------------
[SPARK-30409][SPARK-29173][SQL][TESTS] Use `NoOp` datasource in SQL benchmarks ### What changes were proposed in this pull request? In the PR, I propose to replace `.collect()`, `.count()` and `.foreach(_ => ())` in SQL benchmarks and use the `NoOp` datasource. I added an implicit class to `SqlBasedBenchmark` with the `.noop()` method. It can be used in benchmark like: `ds.noop()`. The last one is unfolded to `ds.write.format("noop").mode(Overwrite).save()`. ### Why are the changes needed? To avoid additional overhead that `collect()` (and other actions) has. For example, `.collect()` has to convert values according to external types and pull data to the driver. This can hide actual performance regressions or improvements of benchmarked operations. ### Does this PR introduce any user-facing change? No ### How was this patch tested? Re-run all modified benchmarks using Amazon EC2. | Item | Description | | ---- | ----| | Region | us-west-2 (Oregon) | | Instance | r3.xlarge (spot instance) | | AMI | ami-06f2f779464715dc5 (ubuntu/images/hvm-ssd/ubuntu-bionic-18.04-amd64-server-20190722.1) | | Java | OpenJDK8/10 | - Run `TPCDSQueryBenchmark` using instructions from the PR #26049 ``` # `spark-tpcds-datagen` needs this. (JDK8) $ git clone https://github.com/apache/spark.git -b branch-2.4 --depth 1 spark-2.4 $ export SPARK_HOME=$PWD $ ./build/mvn clean package -DskipTests # Generate data. (JDK8) $ git clone gitgithub.com:maropu/spark-tpcds-datagen.git $ cd spark-tpcds-datagen/ $ build/mvn clean package $ mkdir -p /data/tpcds $ ./bin/dsdgen --output-location /data/tpcds/s1 // This need `Spark 2.4` ``` - Other benchmarks ran by the script: ``` #!/usr/bin/env python3 import os from sparktestsupport.shellutils import run_cmd benchmarks = [ ['sql/test', 'org.apache.spark.sql.execution.benchmark.AggregateBenchmark'], ['avro/test', 'org.apache.spark.sql.execution.benchmark.AvroReadBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.BloomFilterBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.DataSourceReadBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.DateTimeBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.ExtractBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.FilterPushdownBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.InExpressionBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.IntervalBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.JoinBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.MakeDateTimeBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.MiscBenchmark'], ['hive/test', 'org.apache.spark.sql.execution.benchmark.ObjectHashAggregateExecBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.OrcNestedSchemaPruningBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.OrcV2NestedSchemaPruningBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.ParquetNestedSchemaPruningBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.RangeBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.UDFBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.WideSchemaBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.benchmark.WideTableBenchmark'], ['hive/test', 'org.apache.spark.sql.hive.orc.OrcReadBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.datasources.csv.CSVBenchmark'], ['sql/test', 'org.apache.spark.sql.execution.datasources.json.JsonBenchmark'] ] print('Set SPARK_GENERATE_BENCHMARK_FILES=1') os.environ['SPARK_GENERATE_BENCHMARK_FILES'] = '1' for b in benchmarks: print("Run benchmark: %s" % b[1]) run_cmd(['build/sbt', '%s:runMain %s' % (b[0], b[1])]) ``` Closes #27078 from MaxGekk/noop-in-benchmarks. Lead-authored-by: Maxim Gekk <max.gekk@gmail.com> Co-authored-by: Maxim Gekk <maxim.gekk@databricks.com> Co-authored-by: Dongjoon Hyun <dhyun@apple.com> Signed-off-by: Dongjoon Hyun <dhyun@apple.com>
2020-01-12 16:18:19 -05:00
SQL CSV 13033 13129 136 0.1 12429.1 1.0X
SQL Json 28298 29130 1176 0.0 26987.3 0.5X
SQL Parquet Vectorized 139 151 9 7.5 132.7 93.7X
SQL Parquet MR 314 322 7 3.3 299.5 41.5X
SQL ORC Vectorized 123 143 17 8.5 117.3 106.0X
SQL ORC MR 260 272 9 4.0 248.1 50.1X
[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
Reference in a new issue Copy permalink