When configuring your nodes, do you want them to be part of a single partition or of multiple partitions?
If they will be part of a single partition, just change the multicast address so that nodes can still use the default name for partition: it is easier to manage.
If a given node will be part of multiple partitions, then you need to give a distinct name to each partition and associate, EJB-per-EJB, the configuration to a given partition name.
For each app, all nodes in that partition that the app will be running in should be a single partition. They wont span multiple partitions.
Originally my worry was adding 2 or more partitions with 3 nodes each using the in-memory replication stuff, so that at most 3 nodes worth of replication would occur in any one partition. However, if the sub-node clustering comes into action in the 3.2.x branch, then we can simply use a single partition for one application, and add hundreds of nodes if necessary, so long as only a group of 3 at a time are sub-nodes for state replication. Thus, we can add 3 more nodes each time we want to scale, and sub-node them to each other. Is this the correct understanding of how sub-nodes will work? When will this be ready for testing?