2000-11-11 01:44:01

by H. Peter Anvin

[permalink] [raw]
Subject: Re: sendmail fails to deliver mail with attachments in /var/spool/mqueue

Ralf Baechle wrote:
>
> On Fri, Nov 10, 2000 at 02:18:20PM -0800, H. Peter Anvin wrote:
>
> > Numerically high load averages aren't inherently a bad thing. There
> > isn't anything bad about a system with a loadavg of 20 if it does what
> > it should in the time you'd expect. However, if your daemons start
> > blocking because they assume this number means badness, than that is
> > the problem, not the loadavg in itself.
>
> The problem seems to me that the load figure doesn't express what most
> people seem to expect it to - CPU load.
>

Actually, what most people expect it to represent is schedulability of
new tasks. The problem is more one of:

a) Expecting a fixed relationship between the specific number and the
behaviour of the machine;
b) The long time constants.

On an 8-way machine a load average of 16 is not particularly high, even
if you only count runnable processes, for example.

-hpa

--
<[email protected]> at work, <[email protected]> in private!
"Unix gives you enough rope to shoot yourself in the foot."
http://www.zytor.com/~hpa/puzzle.txt