JMS queue getting slower and slower

Question

We are facing a rather strange behaviour on our WL 8.1.5 cluster (2 nodes).
          One of our JMS queues receives about 400 messages every 5 minutes (the distribution is unbalanced with a peak every 5 minutes). These are forwarded to one MDB instance on every cluster node. Everything works perfect for some time, but there's a creeping performance degradation, although the cluster nodes' resources are far from being exhausted.
          The first message peeks can be processed quite well. After some time, fewer and fewer messages can be processed within a one-minute slot, finally resulting in a queue that gets fuller and fuller. After about 5 hours and some thousand unprocessed (meaning: not processed by the MDB and thus still in the queue) messages the queue becomes a 'storage pipe', forwarding an old message only if the sender inserts a new one.
          Any suggestions ?

Answer 1

I don't think there is any switch to resolve this issue. You need to conduct JMS tuning to improve performance.
          o Start with downstream , Check if your message processing itself is very slow. Monitor Message Driven bean for any rolledback transactions or timed out transactions. Incremental thread dumps during peak processing time may also help.
          o Monitor memory usage and GC frequency of the server during peak condition. If you see high memory usage and high GC frequency, you might want to
          - Tune JVM with Garbage collection and heap size.
          - Turn on message paging to free up memory resources
          o I am not sure, How well it works ? However you might want to try control flow between producer and consumer.
          http://edocs.bea.com/wls/docs81/ConsoleHelp/jms_tuning.html#1121837
          Good luck !!!