ODIn/Website
9
0
Fork 1
Code Pull requests Releases Wiki Activity

Update index.md

Browse source
This commit is contained in:
Oliver Kennedy 2016-08-25 11:59:24 -04:00
parent d52a6a0393
commit 1d803c8555
1 changed files with 37 additions and 0 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

37
src/teaching/cse-662/index.md
View file

@ -96,6 +96,43 @@ After the taking the course, students should be able to:
## Project Ideas
### JITDs
Adaptive indexes are a mechanisms to improve the performance of a database by
modifying indexing structures "on-the-fly" instead of in one large and costly
bulk operation. Adaptive index have been shown to work very well, but only on
worklaods that exhibiti sepcific characteristics. JITDs are a generalization of
adapative indexes, allowing for the expression of many different types of
adaptive indexes. To do this, they provide basic building blocks called nodes
that define index structure. How nodes "fit" together ends up defining index
behavior.
#### Workload-Specific Policies
This project is aimed at coming up with and studying a specific workload and
structuring a policy to manage the creation and usage of nodes based on this
workload. The policy should result in runtime behavoir that is beneficial for
the workload over classic, workload agnostic policies.
### PocketData
While "big data" is all the rage, another important type of data is the data
gathered and curated on your mobile devices. We call this data "pocket data".
#### App Workload Characteristics
In this project you will be given access to anonymized traces of database usage
patterns on mobile devices. This data has been gathered "in the wild" through
the phonelab project. You will be tasked with analyzing this data set and
characterising different applicaitons based on how they utilize pocket data.
Based on your analysis, you will be create a benchmark that exhibits the
characteristics you observed and test the performance of modern mobile
databases and compare and contrast how they perform with respect to classic
database implementations.
#### App Workload Characteristics
### Mimir
Mimir is a probabilistic database for data cleaning. The idea is to
acknowledge that, while automation is amazing and makes data management easy,