[GE users] Job suspend and subordinate queue

Reuti reuti at staff.uni-marburg.de
Fri Jan 13 19:21:16 GMT 2006


Hi Yann,

Am 13.01.2006 um 18:42 schrieb Yann Costes:

> Hi all,
>
> I use Grid Engine 6.0u6 on beowulf cluster made of Linux AMD64  
> computers and I have 2 questions :
> 1) max jobs/user is used on my site. When someone gains its max  
> jobs number, is there a way for this user to suspend on of his job  
> to make running quickly a new job  ? I have made a trial with the  
> command "qmod -sj" but the suspended job still has a "r" status  
> flag and the new job is always in "qw" status.

do you get any feedback after entering qmod -sj <jobid>? It should go  
to state "s" of course. But also suspended jobs are still in the  
system, and will not free up any slot.

> 2) on my site, a queue "seq_long" executing only sequential jobs is  
> subordinate to another queue "para_long" executing parallel jobs.  
> Inside the para_long queue, I have defined two parallel  
> environements : openmp (for OpenMP jobs) and mpi (for MPI jobs). A  
> parallel job using just one host (for OpenMP jobs) always succeeds  
> to make suspend one instance queue of seq_long. However, when a  
> parallel job needs several hosts, no instance queue of seq_long is  
> suspended to make running the parallel job.

This sounds like OpenMP is using all parallel slots on this one  
machine, while MPI is using just one and gathers more from other  
machines. To suspend the serial job with just one slot taken, you  
could change the amount of slots for:

subordinate_list  seq_long=1

(details in man queue_conf). Of course, this might lead to the  
situation using only one parallel slot (which suspends already the  
whole serial queue). Maybe it would be better to use $fill_up as  
allocation rule. Another possibility would otherwise be to have two  
serial slots:

http://gridengine.sunsource.net/servlets/ReadMsg?list=users&msgNo=11735

Cheers - Reuti

>
> Any solution to these problems ?
>
> Thanks in advance.
>
> Yann
>
> ---------------------------------------------------------------------
> To unsubscribe, e-mail: users-unsubscribe at gridengine.sunsource.net
> For additional commands, e-mail: users-help at gridengine.sunsource.net

---------------------------------------------------------------------
To unsubscribe, e-mail: users-unsubscribe at gridengine.sunsource.net
For additional commands, e-mail: users-help at gridengine.sunsource.net




More information about the gridengine-users mailing list