Tuning garbage colletion is really hard. Espescially if you do not know why garbage collection stalls. In general I must say I have never seen a software which shipped with such a good garbage collection configuration as Cassandra.
The thing that looks suspiscious is that the major collections appear regularly in 1 hour intervals. The only software I know of is RMI explicit collections (which I am not certain Cassandra uses) You could avoid those with -XX:+DisableExplicitGC In the Cassandra start script But I assume that Cassandra itself makes use of explicit gc as well so this might have some nasty side effects. Unless someone tells you never to do that on Cassandra I would go for a try and see what happens. Keep flushes and deletions of sstables on your mind though, they are somehow tied to gc I think. Another option (if it is RMI) is to make the RMI full gc threshold larger for timeouts to occur less often. -Dsun.rmi.dgc.client.gcInterval=600000 -Dsun.rmi.dgc.server.gcInterval=600000 The number is ms until a gc is triggered. Anyway.. same warning as before applies Cheers. Von: Rene Kochen [mailto:rene.koc...@emea.schange.com] Gesendet: Montag, 19. Dezember 2011 16:35 An: user@cassandra.apache.org Betreff: Garbage collection freezes cassandra node I recently see the following garbage collect behavior in our performance tests (the attached chart shows the heap-size in MB): [cid:image001.jpg@01CCBE73.8B567F80] During the garbage collections, Cassandra freezes for about ten seconds. I observe the following log entries: "GC for ConcurrentMarkSweep: 11597 ms for 1 collections, 1887933144 used; max is 8550678528" I use Windows Server 2008 and Cassandra 0.7.10 with min and max heap size set to 8 GB. What can I do to make Cassandra not freeze? Just allocate more memory? Thanks, Rene
<<inline: image001.jpg>>
