[ https://issues.apache.org/jira/browse/CASSANDRA-8897?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14543355#comment-14543355 ]
Stefania commented on CASSANDRA-8897: ------------------------------------- [~benedict], this is ready for review. Since my last comment I fixed a few bugs but did not change anything substantial. There was indeed a performance degradation when reading the data generated with the stress profile of CASSANDRA-9240. It was about 600 ops/second slower on my machine (see 9240_test_results.txt attached) and about 1000 ops/second slower on cstar_perf (albeit without a restart). I compared two VisualVM sampler snapshots (attached) and nailed it down to the screenshot also attached. I then changed Ref.GlobalState to use a concurrent hash set rather than a concurrent linked queue and the degradation disappeared. I've also run several cstar_perf tests, with or without the CASSANDRA-9240 profile, with different number of entries and threads. Here are some of them: http://cstar.datastax.com/tests/id/26987156-fa0c-11e4-94a5-42010af0688f http://cstar.datastax.com/tests/id/1915bc2a-fa05-11e4-9b87-42010af0688f http://cstar.datastax.com/tests/id/f7a00bf4-fa09-11e4-9b87-42010af0688f (the diagrams have some problems and occasionally are missing, not sure why) And here are the Jenkins results, which look identical to trunk: http://cassci.datastax.com/view/Dev/view/stef1927/job/stef1927-8897-testall/lastCompletedBuild/testReport/ http://cassci.datastax.com/view/Dev/view/stef1927/job/stef1927-8897-dtest/lastCompletedBuild/testReport/ > Remove FileCacheService, instead pooling the buffers > ---------------------------------------------------- > > Key: CASSANDRA-8897 > URL: https://issues.apache.org/jira/browse/CASSANDRA-8897 > Project: Cassandra > Issue Type: Improvement > Components: Core > Reporter: Benedict > Assignee: Stefania > Fix For: 3.x > > Attachments: 9240_test_results.txt, > snapshot-1431582436640-cpu-backtraces.png, > snapshot-1431582436640-cpu-calltree-compression-8897.nps, > snapshot-1431582436640-cpu-calltree-compression-trunk.nps > > > After CASSANDRA-8893, a RAR will be a very lightweight object and will not > need caching, so we can eliminate this cache entirely. Instead we should have > a pool of buffers that are page-aligned. -- This message was sent by Atlassian JIRA (v6.3.4#6332)