solved problem, thanks.
hosted nodes were different.
On Mon, Jul 23, 2012 at 12:50 PM, mahbube rustaee <[email protected]> wrote:

> $ qconf -sc
> h_vmem              h_vmem     MEMORY      <=      FORCED      YES
> --
> $ qconf -sp mpi30-lb
> pe_name            mpi30-lb
> slots              9999
> user_lists         NONE
> xuser_lists        NONE
> start_proc_args    /opt/gridengine/mpi/startmpi.sh $pe_hostfile
> stop_proc_args     /opt/gridengine/mpi/stopmpi.sh
> allocation_rule    30
> control_slaves     TRUE
> job_is_first_task  TRUE
> urgency_slots      min
> accounting_summary TRUE
> --
> $qconf -sp mpifill
>
> pe_name            mpifill
> slots              9999
> user_lists         NONE
> xuser_lists        NONE
> start_proc_args    /opt/gridengine/mpi/startmpi.sh $pe_hostfile
> stop_proc_args     /opt/gridengine/mpi/stopmpi.sh
> allocation_rule    $fill_up
> control_slaves     TRUE
> job_is_first_task  TRUE
> urgency_slots      min
> accounting_summary TRUE
>
>
> On Mon, Jul 23, 2012 at 12:11 PM, William Hay <[email protected]> wrote:
>
>> On 23 July 2012 08:19, mahbube rustaee <[email protected]> wrote:
>> > Hi,
>> > I defined pe mpi30-lb with allocation_rule 30 and mpifill with
>> > allocation_rule $fillup.
>> > h_vmem is set to real memory.
>> >
>> > I submited a job with sge options:
>> >
>> > #$ -S /bin/bash
>> > #$ -N jobname
>> > #$ -cwd
>> > #$ -l h_vmem=2G
>> > #$ -j y
>> > #$ -pe mpifill 60
>> >
>> > no problem, job will be ran.
>> >
>> > I chenged h_vmem and pe such:
>> >
>> > #$ -l h_vmem=2G
>> > #$ -j y
>> > #$ -pe mpi30-lb 60
>> > job will be deleted immediately(killed) because of h_vmem value.
>> >
>> > Is h_vmem memory request per slot at two above script?
>>
>> That depends on whether you've marked it consumable I believe.
>> >
>> > any hint?
>>
>> Perhaps I'm just being blind but I don't see a difference in the
>> h_vmem requests above.  Given that the only difference is the pe
>> requested then details of the pe config might help us understand.  I'm
>> wondering if something in start_proc_args fo mpi30-1b uses more than
>> 2G h_vmem and this triggers the kill.  Alternatively if the PEs are on
>> different types of node this might cause whatever code you are running
>> to request different amounts of memory.
>>
>> William
>>
>
>
_______________________________________________
users mailing list
[email protected]
https://gridengine.org/mailman/listinfo/users

Reply via email to