We have a JAVA server running on Sun JRE 6u20 on Linux 32-bit (CentOS). We use the Server Hotspot with CMS collector with the following options (I\'ve only provided the relevant
It looks like a GC tuning problem, possibly triggered by a memory leak. I suggest that you add the JVM parameter that turns on GC logging, and see if that gives you any insights.
Does it look like a JVM bug?
Not to me.
When you run out of heap space (for example, as a result of a storage leak) it is common to see the GC running more and more frequently and using a larger and larger percentage of the CPU. There are some JVM parameters that you can set to avoid this, but they have the effect of causing your JVM to keel over sooner with an OOME. (This may sound strange, but it in many cases this is a good thing because it reduces the time until your system is restarted and is able to run normally again.)