Bug? Load avg 2.0 when idle.

Bug? Load avg 2.0 when idle.

[Joerg Sommrey]

Hello,

there is a load average of 2.0+ even if the box is almost idle. (i.e. "top"
shows just one running process: top itself.) Starting two cpu-intensive
processes raises the load average to 4.0+.  How can I determine the
source for the high load, or is this a bug?
I'm running 2.6.9 on a dual-athlon box.

Thanks,
-jo

-- 
-rw-r--r--  1 jo users 63 2004-10-24 19:38 /home/jo/.signature

Re: Bug? Load avg 2.0 when idle.

[Jan Knutar]

On Sunday 24 October 2004 21:29, Joerg Sommrey wrote:
> Hello,
> 
> there is a load average of 2.0+ even if the box is almost idle. (i.e. "top"
> shows just one running process: top itself.) Starting two cpu-intensive
> processes raises the load average to 4.0+.  How can I determine the
> source for the high load, or is this a bug?
> I'm running 2.6.9 on a dual-athlon box.

Look for processes stuck in D state...

Re: Bug? Load avg 2.0 when idle.

[Christian Borntraeger]

Joerg Sommrey wrote:
> there is a load average of 2.0+ even if the box is almost idle. (i.e.
> "top" shows just one running process: top itself.) Starting two
> cpu-intensive processes raises the load average to 4.0+.  How can I
> determine the source for the high load, or is this a bug?
> I'm running 2.6.9 on a dual-athlon box.

Besides other possibilities, a bug in the kernel could be the cause. 
Please check if any process (one or two) is in uninterruptible sleep. 
(using ps axl the state is D)
Furthermore, Magic SysRequest+T (alt-print-t) and the dmesg output could 
give some hints.
If there is nothing suspicious you might try some profiling tool, e.g. 
OProfile. 

There was another bug report about a wrap around load average. I dont know 
if both reports are related. 

cheers

Christian

Re: Bug? Load avg 2.0 when idle.

[Jan Engelhardt]

>> there is a load average of 2.0+ even if the box is almost idle. (i.e.
>> "top" shows just one running process: top itself.) Starting two
>> cpu-intensive processes raises the load average to 4.0+.  How can I
>> determine the source for the high load, or is this a bug?
>> I'm running 2.6.9 on a dual-athlon box.
>
>Besides other possibilities, a bug in the kernel could be the cause.
>Please check if any process (one or two) is in uninterruptible sleep.
>(using ps axl the state is D)
>Furthermore, Magic SysRequest+T (alt-print-t) and the dmesg output could
>give some hints.
>If there is nothing suspicious you might try some profiling tool, e.g.
>OProfile.
>
>There was another bug report about a wrap around load average. I dont know
>if both reports are related.

My ¢2: Start the kernel with init=/bin/sash (or some other minimalistic shell),
if the load average is "normal" (i.e. 0.0), then it's probably somewhere in
userspace.



Jan Engelhardt
-- 
Gesellschaft für Wissenschaftliche Datenverarbeitung
Am Fassberg, 37077 Göttingen, www.gwdg.de

Re: Bug? Load avg 2.0 when idle.

[Joerg Sommrey]

On Sun, Oct 24, 2004 at 08:45:04PM +0200, Christian Borntraeger wrote:
> Joerg Sommrey wrote:
> > there is a load average of 2.0+ even if the box is almost idle. (i.e.
> > "top" shows just one running process: top itself.) Starting two
> > cpu-intensive processes raises the load average to 4.0+.  How can I
> > determine the source for the high load, or is this a bug?
> > I'm running 2.6.9 on a dual-athlon box.
> 
> Besides other possibilities, a bug in the kernel could be the cause. 
> Please check if any process (one or two) is in uninterruptible sleep. 
> (using ps axl the state is D)

On Sun, Oct 24, 2004 at 09:43:42PM +0300, Jan Knutar wrote:
> 
> Look for processes stuck in D state...

Thanks,
there was something hanging in D state.  Strange enough: it was waiting
for modules to be loaded.  After killing a hanging "modprobe" (S state),
the D state processes vanished and load avg returned to normal values.

-jo

-- 
-rw-r--r--  1 jo users 63 2004-10-24 19:38 /home/jo/.signature