[GE users] Parallel job distribution to span subclusters

reuti reuti at staff.uni-marburg.de
Wed Aug 18 14:57:17 BST 2010

    [ The following text is in the "utf-8" character set. ]
    [ Your display is set for the "ISO-8859-10" character set.  ]
    [ Some characters may be displayed incorrectly. ]


Am 18.08.2010 um 13:22 schrieb henry_leyh:

> Hello,
> We have two subclusters, A and B, in our cluster.  The machines in subcluster 
> A are almost identical to those in B, the only difference being slightly 
> faster CPUs.  I can send parallel jobs to either A or B (not caring which) 
> using wildcards in the PE selection.
> What I would like now is a job which doesn't fit in A or B to span over free 
> slots in A _and_ B.  That is, only if there are not enough free slots in 
> neither A nor B it is allowed to combine slots from both A and B.
> Any ideas?  As I understand, sequential numbering of PEs is not available yet?


So you have a third PE which you attached to both hostgroups and can't by reached by accident with the wildcard? The only option I see, is to have a co-scheduler, which will monitor the free slots and use `qalter` for one of the pending jobs to change it from "mpi_*" to "mpi-all" (here a dash) or alike.

Or, instead of altering the job request, change the limit of slots of an RQS of the "limit -pes mpi_all slots=..." rule (here an underscore).

-- Reuti

> Best regards,
> Henry
> -- 
> Henry Leyh ------------ Software Development / System Administration
> Max-Planck-Institut für Plasmaphysik / Stellarator Theory Department
> /*   http://maps.google.com/maps/mm?t=k&ll=54.0743,13.4238&z=19   */
> ------------------------------------------------------
> http://gridengine.sunsource.net/ds/viewMessage.do?dsForumId=38&dsMessageId=275166
> To unsubscribe from this discussion, e-mail: [users-unsubscribe at gridengine.sunsource.net].


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

More information about the gridengine-users mailing list