[torqueusers] Scalability of Torque

Jerry Smith jdsmit at sandia.gov
Thu Apr 23 11:29:42 MDT 2009


> On Thu, 2009-04-23 at 11:05 -0600, Jerry Smith wrote:
>   
>>>   If you do not distinguish between PBSpro and Torque, and whether
>>> they use Maui, MOAB, or pbs_sched as the scheduler. Then #2 and #3
>>> also use PBS or Torque.  #3 is 6400 nodes, #2 is nearly 20,000 nodes.
>>>   
>>>       
>> Does anyone know if each host on these clusters is running a pbs_mom?
>> I know of some systems that have  pbs_mom on the front end ( for job
>> control etc ),  but the compute partitions are running something else.
>>     
>
> On the Cray XT systems running TORQUE to which I have access, there are
> pbs_mom's running on the login nodes and some other service nodes, but
> not on the individual compute nodes.  Access to the compute nodes is
> managed by a Cray-specific service called ALPS, through an mpiexec-like
> command called aprun.
>   
This was my experience as well with the XT3/XT4  system we have here. 
And is why I was asking.   A cluster can be 100k procs, and still only 
have 100 or less pbs_moms running.

Thanks for the follow up Troy.

--Jerry
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://www.supercluster.org/pipermail/torqueusers/attachments/20090423/26e50d14/attachment.html 


More information about the torqueusers mailing list