Open
Description
We often see the current scheduling to result in one worker getting overloaded with a massive queue due to cache hints while other workers are starved for work.
We want the job scheduling algorithm to take cache hints into account, but also to consider worker capacity and availability. There are times where this should be able to really improve our thru put.
Related to, but more general than #168
Metadata
Assignees
Labels
No labels