On Fri, 3 Jan 2003, anindya wrote:
> Hi guys,
>
> I'm new to the world of web polygraph, and I was wondering
> if I could get your help in creating a simple test setup
> for squid. Basically, I'd like to create a benchmark
> to test different builds of squid, for example, with
> async i/o enabled and one without async i/o. I'm looking
> for throughput, response time, and at what point does
> the whole thing just fall over. I'd like to devise a
> test with all uncacheable objects, and perhaps another
> test with a 30% hit rate.
>
> I have only 3 machines at my disposal, the proxy
> squid (solaris), a linux laptop, and a freebsd machine.
> How should I configure the 3 so that will be a "reasonable" test?
> <I realize this is a rather broad question, but I
> don't even understand how to simulate, say 400 users
> hitting the proxy at once>
Hi anindya,
I did something like this last year and made a presentation at OSCON 2002.
You can see the slides from my talk here:
http://www.packet-pushers.com/~wessels/Presentations/OSCON-2002/oscon.html
I used a feature of polygraph called 'rptmstat'. Rptmstat adjusts
the offered load to keep the response time "constant" or within a
certain window. Actually, the polygraph code that I ran had some
tweaks to rptmstat that were never committed to the public polygraph
code. However, you can probably get close to the same results with
the latest version of polygraph.
For some sample rptmstat configuration lines, see this web page:
http://www.squid-cache.org/Benchmarking/tps-vs-diskuse/
Upcoming versions of polygraph will have a feature called 'peakfinder'
which is a better way to find the best performance for a proxy,
subject to certain restrictions (such as constant response time).
Duane W.
This archive was generated by hypermail 2b29 : Mon Feb 06 2006 - 12:00:25 MST