Sun, 12 Dec 2010
/Admin/Apache:
Apache Performance Tuning
I am running my personal server on a rather small Rackspace Cloud machine with 256M of memory.
A default LAMP (plus a couple Python web apps) install seems to occasionally (my websites are
usually not THAT busy) choke and force a reboot. Let's try and avoid that:
First thing I did was install a PHP op code cache called php5-xcache that I have used before. It
is a zero configuration slam-dunk, and my spidy sense suggests that it produces a tangible speed
improvement.
All of the reading I am doing (see references below) prominently mention unloading unused Apache
modules so as to reduce the memory footprint of Apache processes, so this seems like the first place
to start. This is a minimal (not requiring hacking my default Ubuntu server Apache configuration too
much) list of Apache modules I can get by on at the moment:
<![CDATA[
alias.conf cgi.load dir.conf mime.conf php5.load ssl.load wsgi.conf
alias.load deflate.conf dir.load mime.load rewrite.load status.conf wsgi.load
authz_host.load deflate.load headers.load php5.conf ssl.conf status.load
]]>
Now into the murky world of Apache MPM (Multi-Processing Modules). Reading the "Compile-Time
Configuration Issues" section in [4], one is left with the definite impression that the worker MPM
would be a better choice then the prefork MPM in a limited memory situation. So why does prefork
always seem to be installed? A little experiment tells all:
apt-get install apache2-mpm-worker
will result in the removal not only of apache2-mpm-prefork, but also all PHP modules! [5] has the
answer:
"Apache PHP module is reputed to be unstable in multi-threaded environments"
[5] goes on to describe a somewhat complicated procedure for making the multi-threaded
apache2-mpm-worker module work with PHP. Maybe later, if all else fails. Or maybe switch to
lighttpd as a web server. Now back to the stodgy old Apache prefork MPM....
[6] gives some nice details on how one might go about optimizing the prefork MPM configuration.
top is telling me that my Apache processes are currently consuming roughly 30+M per process. My
current default configuration is this:
<![CDATA[
<IfModule mpm_prefork_module>
StartServers 5
MinSpareServers 5
MaxSpareServers 10
MaxClients 150
MaxRequestsPerChild 0
</IfModule>
]]>
which I am going to crank WAY down to this:
<![CDATA[
<IfModule mpm_prefork_module>
StartServers 1
MinSpareServers 1
MaxSpareServers 2
MaxClients 5
MaxRequestsPerChild 200
</IfModule>
]]>
Interesting that the default value of MaxRequestsPerChild was zero. Apparently the point of setting
this value to non-zero is to periodically force Apache processes to be killed and recreated, to combat
memory bloat (aparently they do not release memory?). It is also worth noting that my server's CPU
usage tends to hangout near zero, while memory is always maxed out and swapping is a regular cause of
problems. In top I have seen a single Apache process top out at as high as 50% of memory. I think
killing processes frequently to reduce memory overhead is probably the way to go.
And finally, [6] recommends setting a really small KeepAliveTimeout of 2. And [7] recommends
keeping "Timeout" small, since I have so few processes and do not want them all to be waiting for
a timeout at the same time. So I went for a Timeout of only 10 seconds.
[1] http://wiki.vpslink.com/Low_memory_MySQL_/_Apache_configurations
[2] http://docs.quantact.com/tuning-mysql-apache-for-low-memory
[3] http://groups.drupal.org/node/27174
[4] http://httpd.apache.org/docs/2.0/misc/perf-tuning.html
[5] http://blog.fosketts.net/2010/07/30/high-performance-memory-apache-php-virtual-private-server/
[6] http://www.devside.net/articles/apache-performance-tuning
[7] http://www.360doc.com/content/07/1210/14/15540_883932.shtml
posted at: 21:05 | path: /Admin/Apache | permanent link to this entry
