Hi,
Carsten Leonhardt <[email protected]> writes:
> I still have the issue that pound uses 100% CPU. I'll try to profile it
> in the next days and report back.
so I first tried with gprof which didn't work out at all, probably
because it can't handle multithreaded programs. Next I used sysprof
which shows, if I understand correctly, that the cpu usage comes from
__poll, see below.
__poll is also used by other programs, but there CPU usage is at most
0,04% (Firefox in this case).
Regards,
Carsten
SELF CUMULATIVE FUNCTION
[ 0.00%] [ 96.42%] [/usr/sbin/pound]
[ 3.60%] [ 70.04%] __poll
[ 66.43%] [ 66.43%] - - kernel - -
[ 4.76%] [ 4.77%] __pthread_mutex_unlock_usercnt
[ 2.37%] [ 4.62%] In file /usr/lib/x86_64-linux-gnu/libnanomsg.so.5.1.0
[ 3.20%] [ 3.20%] __pthread_mutex_lock
[ 1.61%] [ 1.61%] __libc_disable_asynccancel
[ 1.49%] [ 1.49%] nn_poll
[ 1.22%] [ 1.22%] __libc_malloc
[ 1.15%] [ 1.16%] __libc_enable_asynccancel
[ 1.01%] [ 1.01%] nn_sock_getopt_inner
[ 1.00%] [ 1.00%] _int_free
[ 0.87%] [ 0.87%] thr_service
[ 0.83%] [ 0.83%] cfree@GLIBC_2.2.5
[ 0.79%] [ 0.80%] nn_getsockopt
[ 0.67%] [ 0.67%] nn_mutex_unlock
[ 0.61%] [ 0.61%] nn_ctx_leave
[ 0.49%] [ 0.49%] nn_mutex_lock
[ 0.31%] [ 0.31%] time
[ 0.29%] [ 0.29%] nn_sock_getopt
[ 0.27%] [ 0.27%] nn_rep_events
[ 0.20%] [ 0.20%] pthread_mutex_unlock
[ 0.19%] [ 0.19%] nn_xrep_events
[ 0.14%] [ 0.14%] In file /usr/sbin/pound
[ 0.11%] [ 0.11%] nn_queue_pop
[ 0.08%] [ 0.08%] __errno_location
[ 0.08%] [ 0.08%] nn_free
[ 0.07%] [ 0.07%] nn_efd_getfd
[ 0.07%] [ 0.07%] nn_sock_rele
[ 0.07%] [ 0.07%] nn_priolist_is_active
[ 0.06%] [ 0.06%] nn_alloc_
[ 0.06%] [ 0.06%] nn_fq_can_recv
[ 0.05%] [ 0.05%] nn_queue_empty
--
pound mailing list
[email protected]
https://admin.hostpoint.ch/mailman/listinfo/pound_apsis.ch
