First of all, I'm biased, so definitely look for more input...
> I should add that entity bean replication is quite
> important as we use entities intensively. I get the
> impression from reading the forum that the clustering
> service is still sort of beta-ish, especially where
> entity replication is concerned.
I would disagree. JBoss Clustering is based on JavaGroups, and JavaGroups (being 5 years old) is quite mature. The JBoss Clustering implementation has also come a long way and is now real stable.
With respect to entity bean clustering, we currently provide support only for shared databases/multiple nodes and commit options B or D.
We're essentially using the database and its transactions to ensure that entity beans are clustered correctly.
I'm working on a replicated transactional cache for JBoss (also based on JavaGroups) which will allow use to (a) provide caching for entity beans (should increase performance significantly) and (b) allow for local DBs (each node has its own database).
Again, I'm biased (being the creator of JavaGroups and lead for the Cache/JBoss project).
I'm also interested in hearing about JBoss Clustering being used in production environments.