Full GC causing application pauses

Question

We have a J2EE application that has been giving us some problems. When the application first starts up it runs fine. Over time we start seeing pauses in the application that seem to be caused by garbage collection pausing the application. We see GC log statements like the following:
[Full GC[Unloading class sun.reflect.GeneratedSerializationConstructorAccessor109]
[Unloading class sun.reflect.GeneratedSerializationConstructorAccessor108]
[Unloading class sun.reflect.GeneratedSerializationConstructorAccessor101]
[Unloading class sun.reflect.GeneratedSerializationConstructorAccessor105]
[Unloading class sun.reflect.GeneratedSerializationConstructorAccessor103]
[Unloading class sun.reflect.GeneratedSerializationConstructorAccessor110]
[Unloading class sun.reflect.GeneratedSerializationConstructorAccessor104]
[Unloading class sun.reflect.GeneratedSerializationConstructorAccessor102]
[Unloading class sun.reflect.GeneratedSerializationConstructorAccessor107]
[Unloading class sun.reflect.GeneratedSerializationConstructorAccessor106]
457267K->266956K(510336K), 3.5484916 secs]
Sometimes the unloading class info is displayed, sometimes it is just the Full GC statment.
We are running WAS 5.1, IBM (Sun) 1.4.1
The app contains EJBs (session, entity and MDB) as well as a JCA adapter for communication with an outside vendor (TCP/IP Sockets)
The app has 500MB allocated for memory. That seems like a lot of memory for this application. Is there anything that could be causing the pauses to increase in frequency and tme? I've checked through the code for hashmap and arraylist handling to make sure we were not holding on to objects once placed into the sets.
Any suggestions would be most appreciated.

Answer 1

There are another options:
1) Buy more memory for your server (it is cheap right?). Larger heap - less GC pause.
2) Update your JVM. Tiger has much better scalability and ergonomics.
3) Setup permgen size.