Question

I have a WorkManager+ThreadPool under Glassfish that uses TIMED_WAIT, such as this:

p: thread-pool-1; w: 2018 (TIMED_WAITING):

at java.lang.Object.wait(Native Method)
at com.sun.corba.ee.impl.orbutil.threadpool.WorkQueueImpl.requestWork(WorkQueueImpl.java:171)
at com.sun.corba.ee.impl.orbutil.threadpool.ThreadPoolImpl$WorkerThread.run(ThreadPoolImpl.java:528)`

Let’s assume for a moment that the threads do nothing (that is, no Work is ever added). Is there a difference in the CPU usage between having 100 threads in the pool and 3000 threads in the pool?

Our testing shows there isn’t, but I’d like to make sure.

Answer 1

I’d say this is OS-specific; we have thousands of threads running in our Java-server softwares on a Linux-server with NPTL (Native Posix Threading Library), but most of them are “sleeping” most of the time, and the cpu-usage has never been really high, usually between 5-20% (on a single 4-core cpu). I couldn’t find any reference right now, but I’m fairly sure I’ve read that with NPTL, the overhead of sleeping threads is negligible.