We've seen problems on our deployed servers running (Intel Xeon) HyperThreading. Unfortunately we do not have one in-house to do much interesting testing with; our solution has been to disable the HT.

Problems we've seen include:
+ reentrancy exceptions ("Reentrant method call detected") in some entities
+ attempted duplicate inserts into session table ("Cannot insert duplicate key row in object 'TBUSERSESSION' with unique index...")

We don't get these problems on our other multi-CPU hosts. Note that we have no clustered deployments either.

We are running 3.2.1 with the default (CMP 2.x) entity cache and lock schemes.

Our app is migrated from WebLogic, where we implemented a JDBC-for-reads and EJB-for-create/update scheme to combat its instance-per-transaction locking scheme (for performance).

Also we're running on Sun JVM 1.4.1_05, using the default CLIENT vm.

Has anyone encountered similar? I've found no mention at Sun or Intel; and scant mention anywhere else. We have NOT to my knowledge experimented with different VM modes - I would like to try the SERVER mode but haven't had the opportunity. Also I wonder if optimistic locking (and perhaps instance-per-tx) would behave differently.

Any thoughts out there? Thanks -
Kent