Re: Another one trying to subscribe :D

From: Henrik Nordstrom <hno@dont-contact.us>
Date: Sun, 13 Nov 2005 10:09:59 +0100 (CET)

Welcome to squid-dev.

On Sat, 12 Nov 2005, Rodrigo [ISO-8859-1] Steinmüller Wanderley wrote:

> My name is Rodrigo Wanderley, I'm just another one working at a ISP ;).
> My interest in squid began few weeks ago when our manager thought it
> would be nice if we had a proxy-cache. I loved the program and started
> reading the code last week. Liked it a lot, and would like to read more
> about how the development is going and the ideas that you guys have.

Thanks!

> I also made a tiny little patch yesterday, and wanted to discuss the
> performance issues of it. The problem was that it was hard to guess the
> exact size to put in maximum_object_size. We would like to cache big
> files from certain sites (like windowsupdate) and let a small default
> for the others. Well, what I did (or tryed to do) was mimic the
> reply-body-max-size option thus adding acl to the maximum-object-size.

Good idea.

If done correctly it should mainly have an effect on performance when
used.

> What I want to do now is profile it and see if putting acls into that
> option is a overkill. Reading the list archives I saw two main options,
> gprof and a patch in squid to enable profiling and view the results in
> the cachemgr. Since the later is not at the code I wander if patching
> the squid for profiling is really the best option...

I'd say profiling this is overkill..

Problem with gprof is the sampling inaccuracy. Most function calls in
Squid is very short in time making the statistical error in run-time
measurements significant. But if you are aware of these limitations of
gprof then it's a valueable tool.

The performance qounters for Squid is an attemp in adding mor accurate
run-time measurements.

Regards
Henrik
Received on Sun Nov 13 2005 - 02:10:03 MST

This archive was generated by hypermail pre-2.1.9 : Thu Dec 01 2005 - 12:00:15 MST