You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@httpd.apache.org by Ian Holsman <ia...@cnet.com> on 2001/09/22 07:23:28 UTC

Listen Q dropouts

Hi.

I'm doing some benchmarking (results @ webperf.org/a2/v25) with the worker MPM
and mod-include.

The file size I'm testing for SSI is ~30K but most of it is in a <if> SSI
so that only 180 bytes are output. (to remove the network bottleneck)

the results look good... the mod-include with no SSI processing is about
as fast as a non-filtered file (with more CPU)

the other surprise (for me) was that The 'acceptmutex Pthread'  gave
better results than the fcntl.

The worker MPM  has improved significantly since last week, and the
mutex contention has been drastically reduced. (mutex-sleeps were between
5-6000, now they range between 2,900-4,000)

I haven't done pre-fork this round.

The next round of benchmarks will be using a web-avalanche box which can simulate
real users more realistically. they should be done next week.

(Thanks to AAron, Sander, Brian, & Justin)

..Ian


Re: Listen Q dropouts

Posted by Justin Erenkrantz <je...@ebuilt.com>.
On Fri, Sep 21, 2001 at 10:23:28PM -0700, Ian Holsman wrote:
> the other surprise (for me) was that The 'acceptmutex Pthread'  gave
> better results than the fcntl.

I still maintain that this should be the default interprocess mutex if 
it is available.  =-)  

In fact, the default - fcntl() - has produced lots of problems for me
on Solaris (as described in STATUS).   -- justin