High load average (~2.0) on an idle PowerPC 64 machine

High load average (~2.0) on an idle PowerPC 64 machine

[Romain Goyet]

I there,

 I've this Quad G5 machine that's sitting pretty much idle with the
latest Debian stable installed, and yet it's got an abnormaly high
load average.
I've detailed the situation over here, maybe you guys will find it
interesting or have something to say :
http://serverfault.com/questions/251299/high-load-average-over-2-0-on-an-idle-machine

The machine doesn't seem to be actually slowed down, it looks more
like an incorrect measurement.

Thank you very much for any help !

 - Romain

Re: High load average (~2.0) on an idle PowerPC 64 machine

[Benjamin Herrenschmidt]

On Thu, 2011-03-24 at 21:23 +0100, Romain Goyet wrote:
> I there,
> 
>  I've this Quad G5 machine that's sitting pretty much idle with the
> latest Debian stable installed, and yet it's got an abnormaly high
> load average.
> I've detailed the situation over here, maybe you guys will find it
> interesting or have something to say :
> http://serverfault.com/questions/251299/high-load-average-over-2-0-on-an-idle-machine
> 
> The machine doesn't seem to be actually slowed down, it looks more
> like an incorrect measurement.
> 
> Thank you very much for any help !

>From memory, this can be due to the thermal control driver's kernel
thread, which essentially does uninterruptible sleeps all the time,
either when waiting for request completion from SMU or i2c.

Since it's pretty much constantly talking to these, it causes an
increase load (they can take time to respond). So in effect it's not
actually hogging the CPU.

I don't know if there's a clean way to fix that.

Now it's possible that there's a different cause, that's just talking
from some vague memories, so some investigations would be useful
regardless.

Cheers,
Ben.

High load average (~2.0) on an idle PowerPC 64 machine

[David Laight]

> I've this Quad G5 machine that's sitting pretty much idle with the
> latest Debian stable installed, and yet it's got an abnormaly high
> load average.

The 'load average' value is rather useless since it seems to
contain any process that is sleeping uninterruptibly, and
IIRC any process that has run at all (for however short a
period) in the current schedule epoch (or whatever period
is relevant).

I can easily generate a linux system that is 99.9% idle
but has a 'load average' of 20 or more.
It would still be 99.9% idle even if the idle time
were based of the actual time outside the scheduler
idle loop (as NetBSD doe) rather than where timer
ticks interrupted.

We also have the related fubar (on x64/amd64) of the
kernel generating stack traces for processes that are
sleeping uninterruptibly for long periods.

	David

Re: High load average (~2.0) on an idle PowerPC 64 machine

[Nathan Lynch]

On Fri, 2011-03-25 at 09:23 +0000, David Laight wrote:
> We also have the related fubar (on x64/amd64) of the
> kernel generating stack traces for processes that are
> sleeping uninterruptibly for long periods.

Turn off CONFIG_DETECT_HUNG_TASK?