httpd-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Marc Slemko <>
Subject Pretty good performance for "" -- PUBNIX (fwd)
Date Mon, 14 Jul 1997 23:52:15 GMT
No, their numbers don't quite jive (1000 hits/second with max of 400
processes... must be small files), but Apache worked for them.  Mildly
suprised no big vendor has got in bed with UUNet yet for their servers.
Or do I recall something about them having done that but it sucked so
much... <g>

---------- Forwarded message ----------
>From: (Piercarlo Grandi)
>Newsgroups: comp.infosystems.www.servers.unix,comp.infosystems.www.misc
>Subject: Pretty good performance for "" -- PUBNIX
>Date: 15 Jul 1997 00:02:12 +0100
>Organization: Home's where my rucksack's
>Message-ID: <>
>X-NNTP-Posting-Host: []
>X-Disclaimer: Contents reflect my personal views only
>Mime-Version: 1.0 (generated by tm-edit 7.106)
>Content-Type: text/plain; charset=US-ASCII
>X-Newsreader: Gnus v5.4.37/XEmacs 19.15
>Lines: 52
>Xref: comp.infosystems.www.servers.unix:31041 comp.infosystems.www.misc:48405

This message is to express, in a totally unsolicited way, my *personal*
(I am in no way officially speaking for my employer here, and I am
posting from my own private machine at home) appreciation of Pubnix's (a
division of UUNET) web hosting services, for they really deserve it...

I have been setting up and running the infrastructure for the
"" unofficial, and highly popular, online publication
about the recent tennis competitions at Wimbledon.

Anticipating great numbers of accesses especially from USA based
browsers the main publication server was chosen to be a dedicated
machine at UUNET, or more precisely Pubnix. This was
"", aka "", aka one of the
two-three machines known as "" (one of them being the
``original'', the other being mirrors).

This unofficial online publication has been a tremendous success, with
several days where we had several million (4-5) hits per day; for
extended periods of time (several hours per day on a few days) we got
well over *a thousand hits per second*, which a total packet traffic
across the main mirror ("") reaching a sustainable
peak of 4,000 packets input or output per second (seriously: four
thousand per second).

When the load started to climb towards these levels UUNET substituted,
without any prompting on our part, and in the course of a few minutes,
the original server (a P100) with a rather beefier one (a PPro 150 with
four times the memory), and were in generally extremely competent and
supportive (and amazed as well at the amount of traffic).

The server was directly attached to a 100Mbits/s FDDI ring, which
connects to the main UUNET backbone; the machine was running BSDI 3.0
and Apache 1.2.0, and set of software tools expertly setup by the
Pubnix/UUNET sysadm team. When I did some load analysis and suggested
some potential improvements (use Squid as server accelerator, etc.),
they responded quickly, as well as when there were problems (process
limits too low: Apache at one point was running as 255 (then with the
limit raised it reached and peaked at 400) processes, given the very
high number of connections).

For the company where I work (MHM Internet) it has been a very
impressive success (I think we outclassed in both quality and
performance and in particular cost/performance the official IBM site at
"", as many appreciative comments we got said), and a
part of it has been the quite impressive sw/hw/network infrastructure at

Another big thanks to the author of the software I used, mostly free
software: the developers of *BSD & BSDI for x86, of Apache, of many
pieces of GNU stuff, of Squid, of mySQL, of "mirror", of Perl and its
many modules, of Java and Kaffe, of "analog", have produced impressive
stuff that has worked well under fairly extreme conditions.

View raw message