[GE users] qrsh and mpi

reuti reuti at staff.uni-marburg.de
Mon Mar 8 20:34:09 GMT 2010


Am 08.03.2010 um 21:21 schrieb bryan:

> When I use qrsh to grab me multiple processors on my cluster, is  
> there an easy way to figure out which nodes gridengine has assigned  
> my job so that I can create a machinefile file for use in mpi?

To create the machinefile is the purpose of start_proc_args in the PE.  
Please have a look at $SGE_ROOT/mpi for some explanations. Depending  
on the MPI implementation you use, this varies though.

Hence the machinefile will be in $TMPDIR, whether it's an interactive  
job or not. The problem might be more, to get all variables set  
automatically as you get a fresh new interactive shell without  
anything inherited from SGE. This depends on the startup mechanism you  
used: -builtin-, rsh, ssh or tigth-ssh.

http://gridengine.sunsource.net/ds/viewMessage.do?dsForumId=38&dsMessageId=242892

-- Reuti

PS: To get a faster output of the used nodes you can use:

$ qstat -g t


> Example:
> When I give the command
>
> qrsh -pe mpi 8
>
> grid engine allocates 8 processors for my job.
>
> Then, to find out which nodes these processors are on, I go to a  
> different terminal and run the command "qstat -f" and note down  
> which nodes have my qlogin job running on them.  I'll then create a  
> machinefile for the mpi job that I want to run from what I learned  
> from the qstat command.
>
> There has to be a better way to get the information and create the  
> machinefile, does anybody know it?
>
> Thanks
>
> Bryan
>

------------------------------------------------------
http://gridengine.sunsource.net/ds/viewMessage.do?dsForumId=38&dsMessageId=247560

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



More information about the gridengine-users mailing list