Re: What a machine for www?

• Mail folder: WWW Talk Apr 94-present
• Next message: Samudra Sengupta: "DA-CLOD version 1.11 now up"
• Previous message: Roy T. Fielding: "Minor patch to wwwstat-1.0 and old2newlog"
• Maybe in reply to: Guenther Fischer: "What a machine for www?"
• Reply: Rex di Bona: "Re: What a machine for www?"

Errors-To: listmaster@www0.cern.ch
Date: Sun, 24 Apr 1994 16:56:47 +0200
Errors-To: listmaster@www0.cern.ch
Message-id: <1786.767199216@hound.cs.indiana.edu>
Errors-To: listmaster@www0.cern.ch
Reply-To: mvanheyn@cs.indiana.edu
Originator: www-talk@info.cern.ch
Sender: www-talk@www0.cern.ch
Precedence: bulk
From: Marc VanHeyningen <mvanheyn@cs.indiana.edu>
To: Multiple recipients of list <www-talk@www0.cern.ch>
Subject: Re: What a machine for www?
X-Listprocessor-Version: 6.0c -- ListProcessor by Anastasios Kotsikonas
Organization: Computer Science Dept, Indiana University
Organization: Computer Science Dept, Indiana University

In article <6945@cernvm.cern.ch> you write:
>Hi,
>our www-machine has much traffic. Sometimes load is about 10. It is a
>Sparc 10 with Solaris. It is also the ftp-Server.
>There are about 10000 GETS per day.

That's about comperable to the transaction count for our server,
except we have a Sparc 2 (x4) running SunOS which also serves FTP and
has a few dozen interactive users, and haven't had load problems
unless lots of people perform extensive searches or somebody runs a
raytracer.

>We have also much CGI's some of them are search engines.

Ah ha...

>We would take another machine, but what is the best for this mix.
>More CPU-Power, more processors, more ... ?

My offhand guess would be fewer execs.  Forking should be dirt cheap
under Solaris (it does copy-on-write, doesn't it?) but CGI requires
exec calls, particularly if you have a lot of frequently used CGIs
which are lightweight and don't involve intensive searching.

You don't mention how much physical memory you currently have; unless
it's really the case that the search engine executing is taking the
lions share of resources (as opposed to the time to load it, the time
to fetch large files from disk, etc) that might be the first candidate
for enhancement.  Our server has 128 megs.

Obviously there are other potential tricks (multiple machines with
round-robin IP, run search engines on another machine via RPCs,
reallocation of nice levels, etc.) depending exactly where the load is
coming from.
--
Marc VanHeyningen  mvanheyn@cs.indiana.edu  MIME, RIPEM & HTTP spoken here