[GE users] mp_load_avg too high WAS: np_load_avg too low

Reuti reuti at staff.uni-marburg.de
Tue Dec 20 20:49:43 GMT 2005


Hi,

the high load also includes processes in D state - D     
Uninterruptible sleep (usually IO).

I got some improvement on one cluster by compiling the kernel for the  
nodes with:

  Processor type and features  --->
  [*] Preemptible Kernel
  [*]   Preempt The Big Kernel Lock

on x86. At SuSE 9.3 times it wasn't suggested for amd64, but maybe it  
changed with the kernel included in 10. - Reuti


Am 20.12.2005 um 21:32 schrieb John Coldrick:

> On Tuesday 20 December 2005 15:19, Rayson Ho wrote:
>> Is your application I/O bound?? What are the numbers reported by
>> "uptime"?? You can also monitor your jobs' processor usage with  
>> "top",
>> and see how often the jobs use the processor.
>>
>> By default, GE only schedules 1 job per CPU...
>
> 	I've been getting odd behaviour with SGE 6.0u6 on SUSE 10 where it's
> reporting my dual x64 Intel system load is 4, 6, 10 even as high as  
> 12, and
> thus not sending jobs to the slots till it "cools down".  Now that you
> mention this, I checked uptime and indeed, I'm getting results like:
>
> 3:29pm  up 35 days  2:58,  5 users,  load average: 8.48, 4.75, 5.73
>
> 	and I've seen that average middle number match what SGE says.  I  
> assume this
> means SGE is determining this(fairly enough) by querying uptime or  
> a similar
> call.  Anyone have any thoughts why uptime would be reporting such  
> high
> numbers?  I'm guessing it's because it's x64, but it seems like a bug
> regardless.  Any thoughts offhand on a workaround?
>
> 	Thanks
>
> 	J.C.
>
> -- 
> John Coldrick                  www.axyzfx.com        Axyz Animation
> 416-504-0425                                         425 Adelaide St W
>                                                      Toronto, ON  
> Canada
> jc at axyzfx.com                                        M5V 1S4
> ---------------------------------------------------------------------- 
> -
> "It's men like him that give the Y chromosome a bad name."
>
> ---------------------------------------------------------------------
> To unsubscribe, e-mail: users-unsubscribe at gridengine.sunsource.net
> For additional commands, e-mail: users-help at gridengine.sunsource.net

---------------------------------------------------------------------
To unsubscribe, e-mail: users-unsubscribe at gridengine.sunsource.net
For additional commands, e-mail: users-help at gridengine.sunsource.net




More information about the gridengine-users mailing list