forgot to mention that we use two Jboss servers in cluster and JBoss messaging database in Oracle 10g.

thanks for the suggestion,let me try that.

Clebert,

I tried with the jmap -histo option which gives me the below results while the production server is running on a normal day. It doesn't look like anything consuming a huge memory compared to the allocated 2GB memory. But our server has become slow already. It takes almost 10 minutes to shut down,however starting the server again is normal. We have found out that after few days if we bring down one of the jboss in cluster and try to start again it cannot join the cluster. or the running one doesn't recognize the one being started.

Size Count Class description
-------------------------------------------------------
16301352 159950 char[]
14233496 15189 int[]
11675784 92177 java.lang.Object[]
11550160 18668 byte[]
11352408 90303 * ConstMethodKlass
6506832 90303 * MethodKlass
5362616 128417 * SymbolKlass
4816200 8454 * ConstantPoolKlass
4815816 200659 java.lang.String
3883456 121358 java.util.TreeMap$Entry
3808368 49192 short[]
3568064 8454 * InstanceKlassKlass
2897120 7281 * ConstantPoolCacheKlass
1658216 17595 java.util.HashMap$Entry[]
1488640 18608 java.lang.reflect.Method
1215312 50638 java.util.HashMap$Entry
1094040 27351 java.util.TreeMap

Thanks for replying. Let us analyze more on the stack trace and the histogram and will update you.

Clebert,

We did the stack trace and found out that the number of WorkerThread for the clients beyond a firewall are steadily increasing by 1 or 2 everyday.It's shwoing up on jconsole as WorkerThread#17[10.**.2*.1**:34128] or similar.When it's reaching about 30-35 it's bringing down the server on it's knees.I was trying to find out if we close the connections properly after each 30 mins refresh period that I mentioned in the first post. We found that the code is correct.
Just for curiosity I tried to find out what happens with standalone clients. So we made a simple standalone client(non-firewall) and connected to jboss and did a stack trace through jconsole while doing load testing. We found out that the worker threads remain even after the client disconnects. Is it normal ?

Our JBoss server in production became very slow so we had to bounce it. But we were able to get some important information. We found this kind of messages in our log file
May 29, 01:25:41 WARN [arjLoggerI18N] [com.arjuna.ats.arjuna.coordinator.BasicAction_58] - Abort of action id ad7d147:95eb:4a1f69f5:7f4 invoked while multiple threads active within it.
May 29, 01:25:41 WARN [arjLoggerI18N] [com.arjuna.ats.arjuna.coordinator.CheckedAction_2] - CheckedAction::check - atomic action ad7d147:95eb:4a1f69f5:7f4 aborting with 1 threads active!

Also we found out that "JMS SessionPool " thread went very high along with the Worker Thread. Please help.

Thanks,
Pratim

Tim,

The number of client session created by remote clients are constant at 9 as they use a connection pool. However we have some local components on the jboss server which opens and closes connections while they put messages to the queues.We are using the java:/JmsXA connection factory. Should we be using the connection pooling mecahnism for local components as well?

Howard,
Nice to see your reply again. As you suggested earlier we are using the jboss-remoting version 2.2.2.SP11

Thanks,
Pratim