Hello All,
We are the GE AssetIntelligence running our J2EE based web application in the famous combination of apache + Jboss servers. Our site was running now almost 2 yrs...after deep JVM memory investigation I found there is huge memory leak in the SMP kernel Linux Jboss systems. When I switched the same machine to run on Non-SMP kernel (i,e EL rather then ELsmp) the memory management handeled very well.
When we run our Jboss application server in SMP kernel we have to bounce our jboss instance minimum 2 times a day to reset the JVM Heap, but with non-SMP kernel its running fine and we dont have to do anything. Is anybody out here facing the same issue?

Note : For JVM heap diagnosis we used JProbe as well as Java inbuilt "jmap" utility to generate the report every hour for heap size and live objects.

I am running following versions of servers,

Hardware :HP DL 380 G4 ( 4 GB RAM , 2 Duel Core CPU 3.60 GHz)
Server 1 : Apache-2.2.11
Server 2,3 & 4 : Jboss-4.0.5.GA

Using apache-2.2.x (mod_loadbalancer) module to connect back end servers.

My JAVA_OPTS vales are

JAVA_OPTS="-Xms3096m -Xmx3096m -XX:MaxPermSize=256m -Djavax.xml.ws.spi.Provider=org.apache.cxf.jaxws.spi.ProviderImpl -Dorg.apache.cxf.bus.factory=org.apache.cxf.bus.CXFBusFactory -XX:+PrintGCTimeStamps -XX:+PrintGCDetails -Dsun.rmi.dgc.client.gcInterval=3600000 -Dsun.rmi.dgc.server.gcInterval=3600000 -Djboss.partition.name=vip-production-lpap03XXXX -Dtsf.environment=PRD -Dnodename=XXXXXXXX"

Any help will be appreciated!!