Slightly off-topic: Has anyone here had much experience with linux v2.5.x/v2.6.0 and running *lots* of threads-per-process? (Assuming a fairly recent version of glibc and the pthreads library)
I ask because of some testing I was doing for an old project I still keep my fingers in. Due to API issues beyond our control, it tends to run into thousands of threads per process. (Which are mainly waiting and thus the switching overhead isn't too bad) Now, under v2.4 linux your memory usage goes up by about your stack size + a little bit more, for each thread you create. Given a good stack of memory, you can run a LOT of threads. So we do, and it all works fine. Under the new pre-v2.6.0 kernels, each thread creation seems to use about 8Mb more in VM - although the resident size is still about what i'd expect. Unfortunately, these poor 32bit machines I'm using don't really like addressing multiple-gigabytes of VM for a single process, and so I'm effectively limited to approx 250 threads - even in simple 'sleep(30);' kind of test routines. Has anyone else hit this, or heard about what is causing it? I've been hunting around for about a week on this one, without finding the cause. (Not having net access at home doesn't help though) -Toby -- Turning and turning in the widening gyre The falcon cannot hear the falconer; Things fall apart, the centre cannot hold; Mere anarchy is loosed upon the world.
