Subject: Re: cpu load
To: None <rmk@rmkhome.com>
From: Andrew Gillham <gillham@vaultron.com>
List: port-i386
Date: 03/01/2002 15:19:38
On Fri, Mar 01, 2002 at 04:03:02PM -0700, Rick Kelly wrote:
> 
> It's not that easy to speculate. For instance, newsyslog could have been
> running. Also, typing 'ls' a few times quickly followed by 'w' will show
> a rise in cpu load, but it falls off in seconds. I just tried this on
> NetBSD/i386, NetBSD/sparc, Solaris 2.6 & 2.8, and SunOS 4.1.1.

11:13AM  up 18:10, 1 user, load averages: 0.54, 0.32, 0.28

Ok, then I'll say that ".28" for a 15 minute interval seems high for
an otherwise idle system.  We'll ignore the 1 & 5 minute averages.

Based on the default crontab newsyslog would have run 13 minutes earlier,
and 'atrun' would have been 3 minutes earlier.  That shouldn't really be
enough to skew the 15 minute average that much. (I would hope)

> Back around NetBSD 1.2 there was a lot of discussion about this.

Sure, but is this a 486 box? :-)
My 1.5.2 celeron machine shows about 0.09 for 15 minute average when idle.
This is at about 15 minutes after so cron should have had a similar impact
on my system.

Admittedly my machine is a 533 celeron so it doesn't directly compare
to the original poster's 333 celeron. 

Anyway, I still think it is high if the box is _not actually doing anything_.
Even with an 'ls -lR' or two my load doesn't go as high.

Running 'top' is likely to provide a better idea of what is going on.

-Andrew