On Sat, 2006-02-11 at 11:09 -0500, John Morrissey wrote:
On Fri, Feb 10, 2006 at 09:14:44PM -0600, Kevin wrote:
On 2/10/06, John Morrissey <jwm@horde.net> wrote:
I'm trying to profile syslog-ng on a set of servers that generate high levels of logging traffic (a few GB/day). Currently, syslog-ng is consuming anywhere from 40-80% of a CPU on a dual Pentium IV Xeon machine.
Even for the slowest possible P4, that seems immensely high. Can you give more information about what on these servers generates the log events, the average log message length, and the average and peak events/second rate?
Can you check if the attached patch decreases your load? It avoids getting back to the poll loop after every log message and loops on the currently processed fd until it either returns EAGAIN or it reaches a hard-coded number. (which is currently 30) Make sure that you touch sources.c.x after applying the patch or the build scripts will require scsh to be installed. If I speculate correctly this will move io_iter lower in your profile. (I would say instead of 70% percent, it will use about 7%) Thanks in advance. -- Bazsi