Well, to make things even more fun we support three db platforms: Sybase, Oracle, and SqlServer. First off, almost all of the deadlock problems were cured after indexing the JBPM tables on all the foreign key columns. (We almost had to call the paramedics for the DBA when we told him we were adding 100 indexes to the database). That tip should be featured more prominently in your documentation. For Sybase, we had to explicitly enable row locking on the tables as well. BTW, indexing also improved 'deleteInstance' speed from 8 seconds per instance to 4 instances per second. Now most often we see 'cannot sync state with session'. This happens when jbpmContext.save() fails for some reason ... we suspect a rollback due to resource contention, but we're not really sure. We occasionally see persistence service errors due to failure to serialize/deserialize an object - we believe these are home-grown variable types (for example, we have java objects to represent tables and picklists). Usually we find an error in our implementation that causes this, but more detailed reporting would be useful. Our system has lots of inter-related moving parts, so creating a trivial example to recreate these issues is pretty much impossible. See http://www.olf.com/ for an overview. Thanx for your help. View the original post : http://www.jboss.org/index.html?module=bb&op=viewtopic&p=4249698#... Reply to the post : http://www.jboss.org/index.html?module=bb&op=posting&mode=reply&a...