The problem seemed to be a kernel issue, as most time spend was in
kernel mode (sys). The machine was exhibiting this behaviour even
in single user mode.
Since the machines weren't in production yet, an upgrade to
PatchKit 4 was done. Which seems to run stable upto now.
Thanks to Dr. Thomas P. Blinn & John Lanier for their insights.
> -----Original Message-----
> I noticed that on 1 node of my 2 node cluster, the load average went
> through the roof at regular intervals. After some investigation, I
> think it is caused by a very slow 'ps' output (since the CAA subsystem
> scripts use 'ps' a lot).
>
>
> On my primary node:
>
> $ time ps -e|wc -l
> 169
>
> real 0m0.099s
> user 0m0.017s
> sys 0m0.084s
>
>
> On the secondary node:
>
> $ time ps -e|wc -l
> 143
>
> real 0m4.999s <<< !
> user 0m0.023s
> sys 0m4.936s
>
>
> Does anybody have any idea why the 'ps -e' on the second node takes
> this long? If I do a plain 'ps', it will return quickly.
>
>
> Any hints, tips?!
>
> Karl
>
>
Received on Mon Aug 09 2004 - 07:27:54 NZST