Review JGroups stacks

Examine the JGroups protocol stacks used in EAP 5 for conformance with known-best practices for the JGroups 2.6.11 code base. Examine any comments for accuracy/completeness as well. Discuss any issues with AS clustering devs.

The stacks can be found at https://svn.jboss.org/repos/jbossas/branches/JBPAPP_5_0/cluster/src/resources/jgroups/jgroups-channelfactory-stacks.xml.

A brief comment on the usage of each stack:

udp – used by HAPartition and all JBC channels (session replication, entity caching, clustered SSO, DistributedState). Which of those uses are significant depends on what the users deploy. So, might have async or sync RPCs, to group or to a single buddy. Kind of a "lowest common denominator" config that should work reasonably for all use cases. Should not be optimized for a particular use case if that highly degrades another use case (e.g. enabling bundling adds a lot of latency to synchronous calls.)

tcp – not used by default. Would be used by default if user set -Djboss.default.jgroups.stack=tcp. Meant to be a "lowest common denominator" equivalent to the "udp" stack for those who want a TCP-based stack.

udp-async – not used by default. Meant to be a tuned config for high volume asynchronous use cases, e.g. JBC REPL_ASYNC.

udp-sync – not used by default. Meant to be a tuned config for high volume synchronous use cases, e.g. JBC REPL_SYNC.

tcp-async, tcp-sync – not used by default. Same usage as udp-async and udp-sync, except TCP-based.

jbm-control, jbm-data – Used by JBM; the latter is expected to be more high-volume, passing JMS messages around the cluster; the former is lower volume control information, e.g. which queues are available on which nodes. The jbm-control stack should use the same shared transport as the "udp" stack to avoid having to configure/isolate a completely separate multicast socket.