Load over 1000

[Poul-Henning Kamp]

In message <Pine.NEB.3.96L.1050221141140.30083I-100000 at fledge.watson.org>, Robe
rt Watson writes:
>
>On Sun, 20 Feb 2005, Christian Jachmann wrote:
>
>> so load over 1024 should not be usual, right?  it should't occur in
>> ordinary practice.  I'm currently testing ULE-scheduler under high load
>> works like a charme. 
>
>Load is a property of the load on a system -- [...]

Actually it is a very misleading indicator of how you use your CPU.

In a world where CPU's were slow enough that even /bin/ls was a CPU
bound task because it sorted the list, loadavg is a good indicator
of how much the CPU is doing.

These days man processes, in particular network servers, never use
a full quantum and therefore the load-average can vastly overestimate
"how much the system has to do".

The thing to remember is that the loadaverage indicates how many
processes would like the CPU, not what or how much they would use
it for once they get access to it.

-- 
Poul-Henning Kamp       | UNIX since Zilog Zeus 3.20
phk at FreeBSD.ORG         | TCP/IP since RFC 956
FreeBSD committer       | BSD since 4.3-tahoe    
Never attribute to malice what can adequately be explained by incompetence.