[ https://issues.apache.org/jira/browse/MAPREDUCE-3641?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13187226#comment-13187226 ]
Vinod Kumar Vavilapalli commented on MAPREDUCE-3641: ---------------------------------------------------- +1. This looks good. I also have an offline confirmation from [~sseth] that this runs fine on a 350 node cluster and improves performance for a few benchmarks. May be good to fix FifoScheduler also to have this, but we can keep that separate. I am going to push this in. > CapacityScheduler should be more conservative assigning off-switch requests > --------------------------------------------------------------------------- > > Key: MAPREDUCE-3641 > URL: https://issues.apache.org/jira/browse/MAPREDUCE-3641 > Project: Hadoop Map/Reduce > Issue Type: Sub-task > Components: mrv2, scheduler > Affects Versions: 0.23.0 > Reporter: Arun C Murthy > Assignee: Arun C Murthy > Priority: Blocker > Attachments: MAPREDUCE-3641.patch, MAPREDUCE-3641.patch, > MAPREDUCE-3641.patch, MAPREDUCE-3641.patch > > > In hadoop-1, the CS is very conservative handing out off-switch assignments, > we need to do the same in YARN. > We noticed performance regressions due to this, particularly for reduces. -- This message is automatically generated by JIRA. If you think it was sent incorrectly, please contact your JIRA administrators: https://issues.apache.org/jira/secure/ContactAdministrators!default.jspa For more information on JIRA, see: http://www.atlassian.com/software/jira