Am 03.04.2012 um 19:14 schrieb Rayson Ho:

> On Tue, Apr 3, 2012 at 12:56 PM, Earl Lazarus <[email protected]> wrote:
>> Bottom line...I need to pre-position some executables on each host without
>> consuming slots.
> 
> How about setting up queues for these kinds of jobs that don't really
> consume much resources?? You can set an arbitrary large number of
> slots for each queue, but with a low job resource limit.
> 
>>  Note that I tried to kick them off in a script of the form:
>> 
>> sppe &
>> exit
>> 
>> but Grid Engine monitors jobs placed in the background and kills them!
> 
> As soon as the job script finishes, Grid Engine kills the job. This is
> the designed behavior.

Exactly, and the same stands for `qrsh -inherit` to start the daemons as child 
of the shepherd. It's possible to get it working like a PVM startup where you 
have unique daemons per job per node, and the daemons continue to run after 
they fork'ed off and still get correct accounting. But it's tricky to implement.

The daemons in your case are also unique per job per node? Do they create any 
daemon information in /tmp or what you assigned as "tmpdir" in the queue 
definition? Can you please describe the "workflow" in case you run your 
application outside of SGE?

-- Reuti


> Rayson
> 
>> 
>> The possibility of avoiding Grid Engine entirely and kicking them off with
>> rsh/ssh is something we want to avoid.
>> 
>> earl
>> 
>> _______________________________________________
>> users mailing list
>> [email protected]
>> https://gridengine.org/mailman/listinfo/users
>> 
> 
> _______________________________________________
> users mailing list
> [email protected]
> https://gridengine.org/mailman/listinfo/users
> 


_______________________________________________
users mailing list
[email protected]
https://gridengine.org/mailman/listinfo/users

Reply via email to