[GE users] how to define the nodes?

Reuti reuti at staff.uni-marburg.de
Thu Jan 17 18:42:45 GMT 2008


Hi,

Am 17.01.2008 um 17:01 schrieb Ryoji Takahashi:

> I am working in the University Tulsa.
> In here, we have 12 quad cores (96 processors) cluster.
> They are just putting Inifiniband with SGE.
>
> I am looking at other sources, etc, however, I cannot find so far any,
> I decide to ask in here.  Probably, someone knows,
>
> so my question is, first, we are try to using mvapich-0.9.9 (OSU)
> for parallel jobs.
>
> in SGE so far, -pe mpich  # processors number
> is there no definition for the nodes?
> , i.e. in PBS, there is -l nodes=#:ppn=#

why are you using this in PBS?

> so we can use mpirun -np #processors number.

Yes, this is the same number as you requested in "-pe mpich  #  
processors number" and you can use "mpirun -np #processors number" in  
your jobscript.

To have it generic you can use "mpirun -np $NSLOTS" in your jobscript.

> In SGE, is there similar definition?  How SGE handle nodes
> and processors?
>
> I was looking this answer, but I didn't find so far any?

In SGE the allocation of number of slots per node is in the PE  
definition. So maybe you need two PEs:

- one with "allocation_rule 1"

- one with "allocation_rule 2"

if youn want to have one or two processes per node for this job.  
Otherwise you can use $round_robin if you don't care.

-- Reuti

---------------------------------------------------------------------
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