On Fri, Jun 26, 2009 at 5:06 PM, Ilan Berkner<iberkner@gmail.com> wrote:
> I haven't done stress testing in a long time... so I'm not sure what the
> best tools are, specifically for our nginx site
Rarely will nginx itself be the bottleneck. For example:
ab testing serving a single image will probably show anywhere from
10,000 - 20,000 req/s, that is because is because after the first
access, it will be in ram and can be served fast and efficiently from
there on. However in a real production environment, where you might be
serving thousands of _different_ files - when the os file cache (in
ram) fills up, those 10,000req/s will become about 300req/s, and in a
worst case scenario (where every request is an os cache miss), it
becomes entirely disk seek bound (which is why SSD hard disk are great
for this sort of job).