Extremely long(10 minute) Full GC times

I have a problem that seems unrelated to traditional GC tuning. Full Garbage Collection on my server (x86, Linux, 2 proc, 2GB memory, 1GB Heap) is taking in excess of 10 minutes. The frequency of the Full GC is about every 8-10 hours. I've combed some posts for issues related to long GC, but long is a relative term, and usually refers to 2-10 second range.

My question is, what kinds of things could cause such an extremely long GC. My theory is that this particular machine is hosting some other applications, most notably a mysql server that is connected to remotely. With this setup, another application could be using resources on this primary server while the Full GC is executing, thus causing a thrashing scenario from a CPU perspective.But even so, it seems that this wouldn't cause Full GC to take 2 orders of magnitude longer.

I've currently changed to the CMS garbage collector, so the long pauses are not a problem, but I'd really like to understand why such a long Full GC pause is even possible, and what can be done about it using the serial Collector

[1093 byte] By [GrubHubMikea] at [2007-11-27 1:31:21]
# 1

Possibly the objects that are being garbage collected, are having their finalize() method called.It's not normally guaranteed that the GC will call an objects finalize() method....but you never know.

In that case, the finalize() method(s) maybe be timing out, eg. trying to close database or network connections, and taking 60 seconds or more to time out individually.

It's just a wild guess...

regards,

Owen

omcgoverna at 2007-7-12 0:34:01 > top of Java-index,Java HotSpot Virtual Machine,Specifications...