Proxy benchmark/tester

On Tue, 9 Dec 1997, Henrik Nordstrom wrote:

> I think the main problem is time. I am spending most of my free time on
> finding the bugs in [Squid]
> ...
> what the development needs right now is some people that seriously
> and systematically tests Squid. Not that much as a general proxy server,
> but how it behaves in strange situations.

Just want to test the following idea that may help with this and quite a
few other things:

How about creating a proxy meter/tester/benchmark? We could create a nice
package that spawns several "virtual" clients and servers and stresses a
given proxy using real (from access logs) and/or generated requests.
Clients and servers could use several hosts to be able to saturate a
proxy. For various purposes, we can have different types of clients and
servers. Each type is responsible for simulating a particular "situation".

The test suite could be used for both performance (benchmarking as in
WebSpec) and testing (debugging) purposes by using appropriate sets of
clients and servers.

Now, let's not spend time describing and discussing thousand possible
things a virtual client/server could simulate (huge/small urls, huge/small
content, fast/slow/jerky connections, reloads, timeouts, various HTTP
control headers, etc., etc.).

Without going into numerous obvious details, how important/useful/urgent
the development of such a test suite is?

Thank you,

Alex.

P.S. As far as I know, similar ideas have been flowing around for quite a
while, but I have not seen any final results yet. There is also a proxy
benchmark being developed at CS department of University of
Wisconsin-Madison, but I doubt it matches our needs.
Received on Tue Dec 09 1997 - 15:24:21 MST