The standard "all" config of JBoss creates 2 partitions, one used for HttpSession replication, the other for lots of other stuff like HA-JNDI, HA-RMI, EJB 2.0 clustering, farming etc.
The jboss.partition.name system property controls the name of this second partition. The name of the partition used for HttpSession replication needs to be configured in the /deploy/tc5-cluster-service.xml file.
We also definitely recommend using a different multicast port or address for different partitions. Otherwise the lower level JGroups protocols need to handle packets for other partitions, which at a minimum degrades performance and sometimes causes anomolous behavior.
That explains what we've been seeing, and may explain some unreliability with replication of deployments made to the farm directory.
I'll make the changes and start testing again
Ok lame question, but related to multicast IP. I want to setup a cluster, but not have my network team freakout about what I am doing and have to explain the whole thing to 3 different managers. Do I need to use UDP or can I just use TCP. To setup this basic cluster and what is the real benefit to using UDP? It isn't a prod environment and just a simple POC.
I say go for multicasting, but set the ip_ttl to a low value so not everyone sees your multicasts.
If you want to use TCP, simply list all members in TCPPING (see sample configs for details).
Regarding the explaining to 3 different managers: watch Office Space and get some good ideas what to do ... :-)