Previously I did some initial setup for JBoss Cache and now I have tested it a little bit on a cluster setup. The setup has two virtual linux images with JBoss 5. In front of them is Apache HTTP for proxying and load balancing.

I started up JBoss on both machines with -c all. This setup has clustering turned on as a default so instances found each other and established JGroup communication channels. Then I deployed the .ear file directly to the deploy directory on each server as the farm deployment did not look too promising. The Apache proxy and balancer were already correctly configured so cookies and sticky sessions worked fine.

The big question is naturally how the performance improved. First I ran a load test without second level caching or query caching in order to have a baseline for comparisons. Then I turned Hibernate second level cache on and executed the same load test. The results were showing around twofold throughput as the average response times halved. I also tried turning on query caching but I did not experience any significant difference with it. Maybe I did not have proper configurations.

I also learned that it is not a good idea to modify data on the DB used by the cluster with Hibernate second level caching on. The nodes communicate sending invalidate messages to each other when data is changed and needs to be refreshed. Modifications outside the cluster mean that the cached data is stale but nodes do not know about it and resulting various errors...