-
Notifications
You must be signed in to change notification settings - Fork 47
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
DLO: Multi-objective optimization for auto-compaction #201
Conversation
...yout/src/main/java/com/linkedin/openhouse/datalayout/ranker/DataLayoutCandidateSelector.java
Outdated
Show resolved
Hide resolved
...alayout/src/main/java/com/linkedin/openhouse/datalayout/ranker/DataLayoutStrategyScorer.java
Outdated
Show resolved
Hide resolved
...java/com/linkedin/openhouse/datalayout/ranker/SimpleWeightedSumDataLayoutStrategyScorer.java
Outdated
Show resolved
Hide resolved
.../src/main/java/com/linkedin/openhouse/datalayout/ranker/BaseDataLayoutCandidateSelector.java
Outdated
Show resolved
Hide resolved
.../src/main/java/com/linkedin/openhouse/datalayout/ranker/TopKDataLayoutCandidateSelector.java
Outdated
Show resolved
Hide resolved
...java/com/linkedin/openhouse/datalayout/ranker/SimpleWeightedSumDataLayoutStrategyScorer.java
Show resolved
Hide resolved
...java/com/linkedin/openhouse/datalayout/ranker/SimpleWeightedSumDataLayoutStrategyScorer.java
Show resolved
Hide resolved
...src/main/java/com/linkedin/openhouse/datalayout/ranker/GreedyMaxBudgetCandidateSelector.java
Show resolved
Hide resolved
.../src/main/java/com/linkedin/openhouse/datalayout/ranker/BaseDataLayoutCandidateSelector.java
Show resolved
Hide resolved
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
lgtm, the only question is why returning indexes instead of strategy objects is important
I feel returning indexes is more intuitive, but I am fine if you think when you incorporate the module in scheduler you can tell it is not a a good idea and we go to list of strategy objects. |
Summary
We plan to develop Auto Compaction for for high ROI tables. Our plan is to treat this as a multi-objective optimization problem by aiming to optimizing two objectives -- maximize file count reduction and minimize compute costs. We score and rank the tables, then choose the top-K tables for each iteration of compaction to remain under allocated compute budget.
This will be used by the job scheduler for candidate selection.
Changes
For all the boxes checked, please include additional details of the changes made in this pull request.
Testing Done
For all the boxes checked, include a detailed description of the testing done for the changes made in this pull request.
Additional Information
For all the boxes checked, include additional details of the changes made in this pull request.