Re: Test the performance of Squid.

Couple of things.

We've already discussed polygraph. That's the obvious choice for
benchmarking web caches. I'm not going to spend any more time on that
issue, because it is so configurable that anything you could ever want
to do, it can be done with polygraph.

Squid can't handle anywhere near 10,000 clients. Not even in an
imaginary workload. Your other two concurrency models will perform
significantly worse, but it may be worth comparing, except that porting
Squid to those two models will be exceedingly difficult for a number of
reasons. (i.e. how do you manage the in memory swap index between
multiple processes/threads? Shared memory or mutexes...)

maer727@sohu.com wrote:
> Hi, pals!
>
> I want to use the following three methods to test the performance of
> Squid. Please give me some suggestions and help.
>
> 1. use one thread for one client, use a process as proxy 2. use one
> process for one client, use a process as proxy 3. use callback method
> in squid, use a process as proxy
>
> I just want to test the respounse time. It has nothing to do with
> cache hit. My purpose is to get some idea of the strength of
> callback. I want to get some test data result. I think test data
> result is always better to talk than pure theory.
>
> I want to write/read socket about a certain size of buffer. I also
> want to generate a C10K problem( I want to generate as least 10K
> clients)
>
> I meet one trouble, I want to know whether there are some tools that
> I can use. Because it will greatly save my time to design the
> framework.
>
> Best regards, George Ma

-- 
Joe Cooper <joe@swelltech.com>
http://www.swelltech.com
Web Caching Appliances and Support