*BSD News Article 93849


Return to BSD News archive

#! rnews 2034 bsd
Path: euryale.cc.adfa.oz.au!newshost.carno.net.au!harbinger.cc.monash.edu.au!news.cs.su.oz.au!metro!metro!munnari.OZ.AU!news.Hawaii.Edu!news.caldera.com!enews.sgi.com!news.mathworks.com!newsfeed.internetmci.com!newsfeed.san.rr.com!usenet
From: bscott@ucsd.edu (Brian Scott)
Newsgroups: comp.unix.bsd.freebsd.misc
Subject: Spikes in server load
Date: Sun, 13 Apr 1997 23:10:43 GMT
Organization: None
Lines: 32
Message-ID: <335165a0.18772346@192.168.0.1>
Reply-To: None
NNTP-Posting-Host: dt6h3n8a.san.rr.com
Mime-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Transfer-Encoding: 7bit
X-Newsreader: Forte Agent .99f/32.275
Xref: euryale.cc.adfa.oz.au comp.unix.bsd.freebsd.misc:39299

Hello.

We've got two FreeBSD servers (2.2.1R and 2.1.7R) on a loaded 10baseT hub. I've
noticed the following behavior: server load is usually 0, but in a single second
will spike to something > 1. Since this is a one-minute average, this would
indicate that there was a _huge_ load spike during the period of one second.
This happens quite often on both computers:

 4:01PM  up 18 days, 19:06, 2 users, load averages: 0.00, 0.02, 0.08
 4:01PM  up 18 days, 19:06, 2 users, load averages: 0.00, 0.02, 0.08
 4:01PM  up 18 days, 19:06, 2 users, load averages: 1.76, 0.39, 0.21
 4:02PM  up 18 days, 19:06, 2 users, load averages: 1.62, 0.38, 0.21
 4:02PM  up 18 days, 19:06, 2 users, load averages: 1.49, 0.37, 0.21

Any ideas on what could be causing this behavior? Or is 'uptime' just
inaccurate? Maybe a software configuration problem? Could this be happening if
the hub is completely maxed out, and FreeBSD is having to trash packets due to
excessive collisions? Netstat reports an average of 50-100 collisions per second
at 50k-100k network traffic per second.

Both servers run Apache 1.2b6, but the web servers aren't heavily loaded. One
runs named; the other does not.

One is a PPro 200, 512k, 64mb, 3Com PCI
The other is a Pentium 200, 256k, 64mb, Intel PCI

Any help would be appreciated.

Thanks,

Brian Scott
bscott@ucsd.edu