[ https://issues.apache.org/jira/browse/CASSANDRA-9345?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]
Alan Boudreault updated CASSANDRA-9345: --------------------------------------- Description: While working on CASSANDRA-7409, we noticed that there were some bad performance with LCS. It seems that the compaction is not triggered as it should. The performance is really bad when using low heap but we can also see that using a big heap there are something wrong with the number of compactions. I can only reproduce this issue in trunk. 2.1 seems OK. Here are the graphs that compare 2.1 vs trunk (Take care when visualizing the graphs, the y-axis scales differ due to some peaks): http://dl.alanb.ca/perf/lcs-2.1-vs-trunk-LH.html http://dl.alanb.ca/perf/lcs-2.1-vs-trunk-BH.html I've attached the stress yaml config used for this test. was: While working on CASSANDRA-7409, we noticed that there were some bad performance with LCS. It seems that the compaction is not triggered as it should. The performance is really bad when using low heap but we can also see that using a big heap there are something wrong with the number of compactions. (Take care when visualizing the graphs, the y-axis scales differ due to some peaks) http://dl.alanb.ca/perf/lcs-2.1-vs-trunk-LH.html http://dl.alanb.ca/perf/lcs-2.1-vs-trunk-BH.html I've attached the stress yaml config used for this test. > LeveledCompactionStrategy bad performance > ----------------------------------------- > > Key: CASSANDRA-9345 > URL: https://issues.apache.org/jira/browse/CASSANDRA-9345 > Project: Cassandra > Issue Type: Bug > Reporter: Alan Boudreault > Assignee: Carl Yeksigian > Fix For: 3.0 > > Attachments: temperature.yaml > > > While working on CASSANDRA-7409, we noticed that there were some bad > performance with LCS. It seems that the compaction is not triggered as it > should. The performance is really bad when using low heap but we can also see > that using a big heap there are something wrong with the number of > compactions. > I can only reproduce this issue in trunk. 2.1 seems OK. Here are the graphs > that compare 2.1 vs trunk (Take care when visualizing the graphs, the y-axis > scales differ due to some peaks): > http://dl.alanb.ca/perf/lcs-2.1-vs-trunk-LH.html > http://dl.alanb.ca/perf/lcs-2.1-vs-trunk-BH.html > I've attached the stress yaml config used for this test. -- This message was sent by Atlassian JIRA (v6.3.4#6332)