powered by Jive Software

Client Session Info Cache diagnosis

I’m running a 2 node Openfire 3.8.2 cluster using the Hazelcast plugin. Watching the caches I noticed that the “Client Session Info Cache” has jumped up over 100% quite a few times, getting up to around 200% at one point. Some other caches have also gone to 100% (vcard and User), but I’ve bumped them up a bit and they’ve settled down now. Whenever something goes over 100% capacity it’s probably a good idea to find out why, or if you can’t, to ask someone nice to explain it. So here I am. Can anyone tell me why Client Session Info Cache would go over 100%, and what I should do to correct it, if anything?

Systems are RHEL 6.4 servers (VMware guests on our internal cloud) with 2 CPU x 8G RAM. Only modification to Openfire startup was to configure Xms=2g, Xmx=2g. Database is MySQL co-located on one of the servers (the db is practically idling, by the way). User accounts are in LDAP. So far memory utilization by Openfire on both servers has stayed below 1G. CPU use is very low on both boxes. We’re using a BigIP LTM to round robin between the servers and each has about the same number of clients connected (right now just under 300 of a possible 6,000+). Conversation activity is light (our high is around 25 at any given time).

Thanks.

Just to update. We’ve upgraded to OpenJDK 1.7.0_45, and my additional params in $OPENFIRE_HOME/bin/openfire now reads:

INSTALL4J_ADD_VM_PARAMS="-server -Xms256m -Xmx2048m -XX:+UseG1GC -Dcom.sun.management.jmxremote"

This has helped our memory utilization enormously (with over 500 users on the system we’re holding steady at under 700M on both boxes – before doing the upgrade we’d topped 1.2G at one point). CPU for the Openfire user is still down around 2% most of the time, occasionally going up to around 25%. But Client Session Info Cache still gets up to around 108% capacity on at least one box. Should I be worrying about that? Note that no one is complaining about performance right now, but then none of us really know what to expect!

Closing the loop on this. Things have been very stable over the last 9 months and have not experienced any performance problems. I’d say our experience is well within normal limits.