On Tue, Dec 14, 2010 at 02:16:34PM -0500, Cathy Guo wrote: > Our application has 9 threads, 1 main thread and 8 monitoring threads. > The main thread does network transactions and the monitoring threads > re-create failed network connections periodically. We ran a test with > broken network connections to see monitoring threads try to re-create > network connections. In this test, we experienced delay in the main > thread. Truss shows the main thread called lwp_park and slept for quite > sometime before a monitoring thread called lwp_unpark to wake it up. The > delay in main thread sometime was more than 1 minute. There's no mutex > contention between these threads.
What is the stack trace for the main thread while it is parked? > The machine I'm using has 4 cpus with 4 cores each. The application only > used less 0.1% CPU. > > I seems to me the machine has enough resources for all 9 threads to run. > I'm puzzled by why the main thread gets parked for so long. threads call lwp_park() when they are waiting for a mutex to be dropped or a condition variable to be fired. The stack trace while it is waiting is going to be the best clue as to what is going on. Cheers, - jonathan _______________________________________________ opensolaris-code mailing list opensolaris-code@opensolaris.org http://mail.opensolaris.org/mailman/listinfo/opensolaris-code
