[ https://issues.apache.org/jira/browse/HBASE-11564?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14073188#comment-14073188 ]
Nicolas Liochon commented on HBASE-11564: ----------------------------------------- Comparing the results with PE reads, 1m rows, 3 replicas and a 100 microseconds delay gives: - w/o the patch: - total: 210s - 95%: 426us - 99%: 550us - with the patch: - total: 174s (i.e. w/o the patch we're 20% slower) - 95%: 208us - 99%: 262us So it's nice. I still have some variation in the results so likely there is still room from improvement somewhere. Commit is on its way... > Improve cancellation management in the rpc layer > ------------------------------------------------ > > Key: HBASE-11564 > URL: https://issues.apache.org/jira/browse/HBASE-11564 > Project: HBase > Issue Type: Bug > Components: Client > Affects Versions: 1.0.0, 2.0.0 > Reporter: Nicolas Liochon > Assignee: Nicolas Liochon > Fix For: 1.0.0, 2.0.0 > > Attachments: 11564.v1.patch, 11564.v2.patch > > > The current client code depends on interrupting the thread for canceling a > request. It's actually possible to rely on a callback in protobuf. > The patch includes as well various performance improvements in replica > management. > On a version before HBASE-11492 the perf was ~35% better. I will redo the > test with the last version. -- This message was sent by Atlassian JIRA (v6.2#6252)