[torquedev] nodes, procs, tpn and ncpus

Ken Nielson knielson at adaptivecomputing.com
Tue Jun 15 14:20:28 MDT 2010


On 06/15/2010 02:09 PM, "Mgr. Šimon Tóth" wrote:
>>>>> The only place where we disagree is that you want to use "procs=X" 
>>>>> where I want
>>>>> to use "nodes=X". I see 2 major downsides: lots of coding work in 
>>>>> torque, and
>>>>> more confusing semantics with mixed (what does "-l 
>>>>> nodes=X,procs=Y" mean?)
>>>>>
>>>>>
>>>> procs=X works today (with Moab, I guess Maui would need to be fixed).
>>>> No changes needed for TORQUE, except that TORQUE doesn't know what
>>>> procs means, it just runs the job on the nodes Moab tells it too...
>>>>
>>>> so if you don't care about it not working without Moab, we have this
>>>> solution today, and it works correctly with EXACTNODE.
>>>>
>>>> In my opinion, I think nodes=X,procx=Y should generate a syntax 
>>>> error...
>>>> _______________________________________________
>>>>
>>>>
>>> I have now changed my mind and agree with Glen that this is an 
>>> error. My
>>> reasoning is that to try and add it to the node spec as in -l
>>> nodes=2:ppn=4+procs=2 is redundant. It is the same as -l nodes=2:ppn=4
>>> -l procs=2. The later is already supported by TORQUE. TORQUE also 
>>> throws
>>> an error if the -l nodes=2:ppn=4+procs=2 syntax is used.
>>>
>>>    From a coding point of view it is much easier to support the current
>>> syntax. We just now need to build in the ability to parse the -l 
>>> procs=x
>>> and translate it into usable processors for the job.
>>>
>>>
>> Only one procs resource is allowed per qsub. Currently, if you do
>> something like qsub -l procs=4 -l nodes=3 -l procs=2 the last procs
>> statement is the one that gets used. So in this instance the
>> Resource_List.procs element will be equal to 2.
>>
>> Just wanted to post the behavior.
>
> That should be true for any resource passed through -l resource=value.
>
I have been looking at this parsing problem from inside of TORQUE where 
it is difficult to combine the concept of node and processors with the 
current routines. But we could have qsub map the -l nodes=x+procs=y to 
be -l nodes=x -l procs=y. That way we can keep the current internal 
behavior of TORQUE and allow users the new syntax. David Beer did this 
with -P option for qsub in 2.4.8 for groups.

Ken




More information about the torquedev mailing list