Re: Squid Throughput/Performance Testing

On Tue, 2008-04-08 at 13:57 +0100, Bradley Kite wrote:

> I'm interested in the performance of squid, and in particular, how do
> the developers optimise/benchmark/stress-test the code - I'd like to
> use similar ideas for evaluation.
>
> Any hints/tips that you can recomend on a low budget? I've looked at
> Web Polygraph, and in a lab environment this seems excellent, but I
> think it requires more hardware than I have available to me.

Polygraph is faster than Squid so it is unlikely that you need a lot of
hardware for a realistic benchmark unless you have a cluster of high-end
Squid boxes. A couple of comparable PCs is usually sufficient to test a
single Squid box. If you go above 700 request/second rates on standard
workloads, use Gigabit NICs.

Latest standard workloads have How To documents that you can use to set
things up. The most difficult thing is often increasing the number of
filedescriptors available to Polygraph, but you can usually find
OS-specific instructions on the web (and you will need that for Squid
anyway).

Depending on your request rates, you may need to change the standard
workloads a little to increase the allowed per-host rate. There are a
few examples on how to do that in the Users mailing list archives. Other
shortcuts (like disabling DNS part of the tests) are possible. Post to
Users if you get stuck.

Good luck,

Alex.
P.S. Old cache-off results did use a lot of low-end PCs on most benches,
but PCs are a lot faster these days and GBit NICs help.
Received on Tue Apr 08 2008 - 08:45:18 MDT