[GE users] np_load_avg and thresholds

Shannon V. Davidson svdavidson at swbell.net
Tue Aug 31 23:02:30 BST 2004


Oops - my bad.  I see that np_load_avg is actually calculated in the 
exec daemon and uses the load value nproc just retrieved from the 
kernel, so it doesn't help to override it in the qmaster.  It might work 
to override it in a load sensor.

Shannon


Shannon V. Davidson wrote:

>
> Setting num_proc in the execution host's complex_values seems to work 
> for me:
>
>
> $ qhost
> HOSTNAME                ARCH         NCPU  LOAD  MEMTOT  MEMUSE  
> SWAPTO  SWAPUS
> -------------------------------------------------------------------------------
> global                  -               -     -       -       -       
> -       -
> appro                   lx24-amd64   2.00  0.99    1.9G  163.4M    
> 1.9G     0.0
> pogo                    lx24-x86     2.00  0.32    2.0G  351.9M    
> 3.9G   70.7M
>
> $ qconf -mattr exechost complex_values num_proc=1 appro
>
> $ qhost
> HOSTNAME                ARCH         NCPU  LOAD  MEMTOT  MEMUSE  
> SWAPTO  SWAPUS
> -------------------------------------------------------------------------------
> global                  -               -     -       -       -       
> -       -
> appro                   lx24-amd64   1.00  0.99    1.9G  163.2M    
> 1.9G     0.0
> pogo                    lx24-x86     2.00  0.25    2.0G  353.7M    
> 3.9G   70.7M
>
>
> Cheers,
> Shannon
>
>
> Jay Lessert wrote:
>
>>On Tue, Aug 31, 2004 at 04:12:21PM -0400, Don Shesnicky wrote:
>>  
>>
>>> 
>>>    
>>>
>>>>>You can, of course, set the processor count manually in your SGE
>>>>>config if you decide to keep HT on but want to only account for the
>>>>>        
>>>>>
>>>'real' CPUs in your SGE config.
>>>
>>>    
>>>
>>>>This is apart of an old thread on processor counts, xeon processors
>>>>and the number of slots.  How do I set the processor count for an
>>>>individual host?
>>>>      
>>>>
>>
>>I saw the same old thread you did, read the docs, and came to the
>>conclusion that according to the docs, you CANNOT edit num_proc
>>or processors.  The host_conf(5) man page is pretty explicit about
>>this, I'd be happy to hear otherwise.
>>
>>I've got a small bank of dual-Xeon Sun V60x's I just installed.  I
>>wound up just doing:
>>
>>    qconf -rattr queue load_thresholds load_short=1.50 $list_of_v60x_queues
>>    qconf -mattr queue slots 2 $list_of_v60x_queues
>>
>>...in my post-install script.
>>
>>I did leave HT off (BIOS setting) on one of the boxes.  Turns out I
>>like that better, anyway.  For my job mix (VCS, Spectre, Nanosim,
>>DC/PC/Astro) these boxes behave better with load_avg=2.0 if HT is
>>off. (This is RHEL3, stock 2.4.21-4.ELsmp kernel.)
>>
>>So after tapeout when I get a chance I'll disable HT on all of 'em,
>>and the problem goes away.
>>
>>  
>>
>
>
>-- 
>___________________________________________
>
>Shannon V. Davidson <svdavidson at swbell.net>
>Senior Software Engineer           Raytheon
>636-479-7465 office        443-383-0331 fax
>___________________________________________
>
>  
>


-- 
___________________________________________

Shannon V. Davidson <svdavidson at swbell.net>
Senior Software Engineer           Raytheon
636-479-7465 office        443-383-0331 fax
___________________________________________





More information about the gridengine-users mailing list