You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Describe the bug
In order to fully utilize the benefits of multithreaded shuffle, the first GPU run (when autotuner runs on cpu logs) should not increase partitions to values > 200.
We might have to fork this logic for GPU iterations where the benefit of higher partitions overrides the ones from MT shuffle.
The text was updated successfully, but these errors were encountered:
Scope: update auto-tuner to detect that it is processing a CPU event log and then default shuffle partitions to 200. When GPU event log is detected, it can keep existing logic.
Describe the bug
In order to fully utilize the benefits of multithreaded shuffle, the first GPU run (when autotuner runs on cpu logs) should not increase partitions to values > 200.
We might have to fork this logic for GPU iterations where the benefit of higher partitions overrides the ones from MT shuffle.
The text was updated successfully, but these errors were encountered: