Re: COSS stuff

----- Original Message -----
From: <adrian@creative.net.au>
To: <squid-dev@ircache.net>
Sent: Monday, February 14, 2000 5:50 PM
Subject: Re: COSS stuff

> On Mon, Feb 14, 2000, Henrik Nordstrom wrote:
> > I am planning on taking a bash at it shortly in the limited environment
> > I have at home. I will keep you posted.
> >
> > I do not have much left to do on my 2.2.STABLE5 branch now, only a
> > sporadic segfault which hasn't been tracked down yet. Quite hard when
> > Linux does not allow coredumps from threaded applications, and the
> > automatic stack-trace functionality in GNU libc apparently does not work
> > on Alpha processors....
> >
> > Once we have written some documentation on our administative interface
> > and some other minor stuff we will begin playing around with "the next
> > generation", where our initial goal for the Squid part is to double the
> > performance from current ~10Mbit of live incoming internet traffic to at
> > least ~20Mbit per box using the same hardware configuration, and
> > hopefully also allow a larger cache size with the same amount of main
> > memory. Timeframe is set to be able to deliver within 6 monts.
>
> I have a similar aim, however squid has a lot of 'cruft' in its design
> philosophy which currently limits its top speed. Its my hope to keep
> pushing forward enough with my fs/network code modifications (including
> kernel stuff) until some other interested parties take it over, so
> I can work on other cache related stuff.

Really? I don't see anything in the squid design that I would say limits its
top speed. I have a feeling the speed issues just have to do with code
optimization. There are probably some core routines that are alot slower
than they could be. (My gut tells me the HTTP header routines are probably
terribly slow/inefficient, but thats just a feeling, I haven't actually
looked yet.)

Eric Stern - Senior Product Developer
Industrial Code & Logic Inc. - (519) 249-0508
http://www.indcl.com
--------
WebSpeedWare - web caching doesn't have to be expensive!
Received on Tue Feb 15 2000 - 11:01:45 MST