Return to BSD News archive
Path: euryale.cc.adfa.oz.au!newshost.carno.net.au!harbinger.cc.monash.edu.au!munnari.OZ.AU!news.mel.connect.com.au!news.mira.net.au!inquo!in-news.erinet.com!ddsw1!news.mcs.net!www.nntp.primenet.com!nntp.primenet.com!news.sgi.com!news1.best.com!nntp1.best.com!flash.noc.best.net!not-for-mail
From: dillon@best.com (Matthew Dillon)
Newsgroups: comp.os.linux.misc,comp.unix.bsd.freebsd.misc,comp.infosystems.www.misc
Subject: Re: How many hits a second on a unix server?
Date: 25 Sep 1996 01:03:45 -0700
Organization: Best Internet Communications, Inc. (info@best.com)
Lines: 67
Distribution: world
Message-ID: <52ap11$9iv@flash.noc.best.net>
References: <323ED0BD.222CA97F@pobox.com> <51nn4m$gn3@usenet.srv.cis.pitt.edu> <51si5a$79o@magic.metawire.com> <Dy1FJp.n21@zorch.SF-Bay.ORG>
NNTP-Posting-Host: flash.noc.best.net
Xref: euryale.cc.adfa.oz.au comp.os.linux.misc:131485 comp.unix.bsd.freebsd.misc:27916 comp.infosystems.www.misc:44060
:In article <Dy1FJp.n21@zorch.SF-Bay.ORG>,
:Scott Hazen Mueller <scott@zorch.sf-bay.org> wrote:
:>>300 hits/s = 25,920,000 hits/day, more than Netscape,
:>
:>Netscape took 100 million one day last week.
:>
:>However, they've got several roomsful of servers.
:>
:>Supposedly, according to Bill Foss, they got about 60-80 hps from
:>Sparc/Solaris 2.x and 100-120 hps from SGI Indigo/IRIX 5(?). Assuming a flat
:>load distribution, those work out to 5184000, 6912000, 8640000 and 1036800
:>hits per day respectively. However, on my site what I call the "peak load
:>factor", i.e. the value by which the peak load exceeds the average, is ~2, so
:>those figures would have to be cut in half to represent a realistic view of
:>what a live server will do in production.
:>
:>And, just to throw a joker into the deck, Matt Dillon reported yesterday that
:>he was getting 70 hps from an SGI Challenge L, which puts those SGI Indigo
:>figures in doubt.
:>
:> \scott
I should annotate my comment, however... we are serving over 16,000 user's
web pages off of that challenge L. There are scaling and caching issues
that create severe limitations due to the design of IRIX... filesystem
locking issues, network protocol stack issues, directory size issues,
buffer caching issues, etc... The machine is, literally, serving
over 44 GBytes worth of files, with over 40 GBytes/day in transfers.
Also, we allow CGI and are running the ftp server on the same machine.
Even so, we expected the machine to be able to do greater then 200 hits
a second and we are only getting 70 or so out of it.
A long, long time ago we were running our web stuff off of FreeBSD... this was
right at the beginning. We were doing about 45 hits/sec on a pentium-90.
We are actually starting to move back to FreeBSD boxes again.. this time on
200 MHz pentium pro boxes. If I were to hazzard a guess, taking into account
upgrades made to the www server inbetween then and now, and the increased
cpu clock, I would say a pentium-pro 200 PC running FreeBSD with sufficient
memory and networking (i.e. 100Base-T ethernet) should be able to do
100 hits/sec with the kind of load we deal with.
If you are serving only your own pages, or pages which you have control
over, you will get much better results then what we were/are/will-be
getting. For a limited, fully cacheable set of pages, no CGI, and
nothign else running (i.e. no ftpd), you ought to be able to do
150 hits/sec on a pentium-pro 200.
I should note that once you get over 40 hits/sec, you have to start worrying
about your network. If you are using 10Base-T, you have to throw it into
an etherswitch rather then a hub. Once you get over 60 hits/sec, you will
likely max out a 10MBps ethernet even if it is on a private segment...
at that point you need to move to a 100 MBps ethernet or play with
multiple 10Mbps ethernets (you can actually *do* 1 MByte/sec on an ethernet
if you have two ethernets and use one for incoming and one for outgoing
traffic to remove the collision factor completely). FDDI is also
an option, but a more expensive one because there are no cheap FDDI switches.
Linux would also make a reasonable web server, but I haven't worked on linux
for at least a year so I cannot comment on its current-day efficiency vs
FreeBSD.
-Matt
--
Matthew Dillon Engineering, BEST Internet Communications, Inc.
<dillon@best.net>
[always include a portion of the original email in any response!]