[torqueusers] only 1 core jobs in the queue

Steve Young chemadm at hamilton.edu
Tue Mar 3 07:26:42 MST 2009


On Mar 3, 2009, at 3:02 AM, web master wrote:

> Hi
>
> 2009/3/2 Garrick Staples <garrick at usc.edu>:
>> On Mon, Mar 02, 2009 at 01:54:07PM +0100, web master alleged:
>>> Hi
>>>
>>> I am very inexperienced with torque, and found myself trying to  
>>> set up
>>> a queueing system on a // cluster, with 12 8-core nodes.
>>> Majority of users only needs to run single processor jobs, so I  
>>> created
>>> a queue for them and a queue for // jobs, i.e. more than 1 core.
>>>
>>> Fast queue specs:
>>> # Create and define queue fast
>>> #
>>> create queue fast
>>> set queue fast queue_type = Execution
>>> set queue fast max_running = 24
>>> set queue fast resources_max.ncpus = 1
>>> set queue fast resources_max.nodect = 1
>>> set queue fast resources_max.nodes = nodes=1:ppn=1
>>> set queue fast resources_max.size = 1
>>> set queue fast resources_max.walltime = 12:00:00
>>> set queue fast resources_default.ncpus = 1
>>> set queue fast resources_default.nodes = nodes=1:ppn=1
>>> set queue fast resources_default.walltime = 12:00:00
>>> set queue fast enabled = True
>>> set queue fast started = True
>>>
>>> However, I cannot manage to stop jobs using up to 4 cores to be  
>>> enqueued in
>>> the fast queue.
>>>
>>> XXXXX at nodo0>qsub -q fast -l nodes=1:ppn=1 launch_single.sh
>>> 11002.nodo-ha
>>> XXXXX at nodo0> showq | grep MYUSERNAME
>>> 11002              MYUSERNAME    Running     1    12:00:00  Mon  
>>> Mar  2 14:23:19
>>> giupponi at nodo0> qsub -q fast -l nodes=1:ppn=2 launch_single.sh
>>> 11003.nodo-ha
>>> XXXXX at nodo0> showq | grep MYUSERNAME
>>> 11003               MYUSERNAME   Running     2    12:00:00  Mon  
>>> Mar  2 14:23:30
>>> giupponi at nodo0:~/Vesicles/pka_outside/TestQueues> qsub -q fast -l
>>> nodes=1:ppn=9 launch_single.sh
>>> qsub: Job exceeds queue resource limits MSG=cannot locate feasible  
>>> nodes
>>>
>>> launch_single.sh
>>>
>>> #!/bin/bash
>>> #PBS -N sleep
>>> #PBS -e sleep.e
>>> #PBS -o sleep.o
>>>
>>> sleep 10
>>>
>>> Assuming I am not doing anything wrong, how can I obtain the  
>>> desired behaviour?
>>> If not with torque, with maui maybe?
>>
>> Torque only counts nodes, e.g. resources_max.nodect.  nodes=1:ppn=1,
>> nodes=1:ppn=2, and nodes=1:ppn=9 are all 1 node.
>
> I understand.
> But I thought that
> set queue fast resources_max.nodes = nodes=1:ppn=1
>


I'm not sure I have any good answers for you but it is my  
understanding that this should be a number (1 in your case) not a  
string (nodes=1:ppn=1). Did you try this:
set queue fast resources_max.nodes = 1? The ncpus setting before it  
should constrain you to one cpu. I could be wrong here but at first  
glance that is what I was thinking. Hope this helps.




> would not allow 2-core jobs to be assigned to that queue.
>> From what you say, it seems  there is no way to route to different  
>> queues
> different jobs that need a different number of cores
> Is that so?
>
> Cheers
> _______________________________________________
> torqueusers mailing list
> torqueusers at supercluster.org
> http://www.supercluster.org/mailman/listinfo/torqueusers



More information about the torqueusers mailing list