[Dovecot] High cpu usage OpenSolaris
All,
Nevermind... I discovered the problem. It was in the Solaris Sun Studio 12 compiler. I created a debug version to begin testing, but that version worked without sucking up the cpu. So it looks like the optimizer messed up. If I used the -fast optimization, it sucks up cpu, but with -xO5 all is well.
Gary
Hi,
I've built 64-bit dovecot 1.0.2 on Opensolaris to replace version 1.0.0rc10. It functions fine, but when I connect, it eats up all available cpu. My guess is that the poll thread is running continuously. Any ideas on what I should do to verify this or does anyone know of a fix?
Thanks, Gary
On Sun, 2007-07-29 at 21:27 -0400, Gary Gendel wrote:
All,
Nevermind... I discovered the problem. It was in the Solaris Sun Studio 12 compiler. I created a debug version to begin testing, but that version worked without sucking up the cpu. So it looks like the optimizer messed up. If I used the -fast optimization, it sucks up cpu, but with -xO5 all is well.
I heard about this before too. Would be nice to know if it really is a bug in Sun Studio or if the problem is caused by something done wrong in Dovecot's code.
On 8/1/07, Timo Sirainen tss@iki.fi wrote:
On Sun, 2007-07-29 at 21:27 -0400, Gary Gendel wrote:
All,
Nevermind... I discovered the problem. It was in the Solaris Sun Studio 12 compiler. I created a debug version to begin testing, but that version worked without sucking up the cpu. So it looks like the optimizer messed up. If I used the -fast optimization, it sucks up cpu, but with -xO5 all is well.
I heard about this before too. Would be nice to know if it really is a bug in Sun Studio or if the problem is caused by something done wrong in Dovecot's code.
can't you use dtrace or something to know where the CPU time will go ?
-- DINH Viêt Hoà
On Wed, 2007-08-01 at 15:12 +0200, DINH Viêt Hoà wrote:
On 8/1/07, Timo Sirainen tss@iki.fi wrote:
On Sun, 2007-07-29 at 21:27 -0400, Gary Gendel wrote:
All,
Nevermind... I discovered the problem. It was in the Solaris Sun Studio 12 compiler. I created a debug version to begin testing, but that version worked without sucking up the cpu. So it looks like the optimizer messed up. If I used the -fast optimization, it sucks up cpu, but with -xO5 all is well.
I heard about this before too. Would be nice to know if it really is a bug in Sun Studio or if the problem is caused by something done wrong in Dovecot's code.
can't you use dtrace or something to know where the CPU time will go ?
Last I heard it was that poll() was called all the time with zero (or 1ms?) timeout. So the problem is probably in src/lib/ioloop.c io_loop_get_wait_time() getting optimized wrong.
Last I heard it was that poll() was called all the time with zero (or 1ms?) timeout. So the problem is probably in src/lib/ioloop.c io_loop_get_wait_time() getting optimized wrong.
Hmm.. poll/timeout problems. Sounds suspiciously like:
Just a wild guess, but hey... :-)
- Peter
participants (4)
-
DINH Viêt Hoà
-
Gary Gendel
-
Peter Eriksson
-
Timo Sirainen