[GE users] consumable resources in a pe job

aja alena.plestilova at sun.com
Thu Jun 4 13:26:21 BST 2009

Hi Andreas,
what version of Grid Engine are you running? Since 6.2. version you can 
define complex values consumable per job, see the man complex(5), 
consumable section.
Hope this helps,

On 06/03/09 15:44, ah_sunsource wrote:
> Hi *,
> SGE assigns consumable resources per parallel task for a PE job. We have
> set up a PE called "multicore" that should be used by our users if their
> job is using more than one cpu on a single host (without mpi, etc.). It
> is set up like this:
> [oreade38] ~ % qconf -sp multicore
> pe_name           multicore
> slots             256
> user_lists        NONE
> xuser_lists       NONE
> start_proc_args   /bin/true
> stop_proc_args    /bin/true
> allocation_rule   $pe_slots
> control_slaves    FALSE
> job_is_first_task FALSE
> urgency_slots     min
> One of our user's jobs needs 10GB of RAM regardless how many parallel
> tasks it is using. He is submitting his job like this:
> qsub -l h_vmem=10G -pe multicore 2-8 jobscript.sh
> This leads to the situation that 20GB are assigned to the job if it gets
> two slots and 80GB if it gets eight slots - this is not what he wants
> (we even do not have hosts with 80GB memory). When using a static value
> for number of pe slots you could simply calculate the number to be
> requested for h_vmem, but in this case it is simply impossible...
> Is there a way to declare h_vmem static to the whole job instead of
> assigning h_vmem to any slot?
> Cheers & Thanks
> Andreas

aja & sun
e-mail: aja at sun.com


To unsubscribe from this discussion, e-mail: [users-unsubscribe at gridengine.sunsource.net].

More information about the gridengine-users mailing list