leixm commented on PR #307: URL: https://github.com/apache/incubator-uniffle/pull/307#issuecomment-1308294460
> We should allocate continuous n reduce partitions to one shuffle server to avoid that too many task read the same shuffle server. If we have 700 reduce tasks, our concurrency is 100, we usually start 0 - 99 task first. if we allocate 10 continuous reduce partitions to one shuffle server, shuffle server 0 will have [0, 9] [80, 89] reduce partitions, shuffle server 1 will have [ 10, 19], [90, 99] reduce partition, shuffle server 3 will have [20, 29],[100, 109] reduce partition .... We also need some performance test. How to decide how many contiguous partitions need to be allocated, like `10` mentioned above. -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected] --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
