Hi Jessica,

Normally you should not have to install applications on compute nodes but
you should use head node as an application server. We for example do that
by installing all the applications under /local/ClusterApps directory on
the head node and then we NFS export this directory and mount it on the
remote clients. The application environment we manage using modules please
check out this website http://modules.sourceforge.net/. When a job starts
on the compute nodes job script loads needed environment modules which then
set up all the necessary paths and libraries for the applications. this is
how it is usually done on the Linux clusters.

Best regards,

Wojciech

On 11 October 2012 23:17, Andy Riebs <[email protected]> wrote:

>  SLURM expects to find an application's code and libraries on each node
> being used. It does not attempt to download or stage the application before
> execution.
>
> Andy
>
>
> On 10/11/2012 05:21 PM, Jessica Rabbet wrote:
>
> We have a new cluster. All of our applications are on the head node. When
> submitting a job to slurm, the job fails stating it can't find the
> application or libraries. Do the applications need to be installed on the
> compute nodes as well? Or is there some way to make slurm use the
> application on the head node before sending the data to the compute nodes?
>
>
> --
> Andy Riebs
> Hewlett-Packard Company
> High Performance Computing+1-786-263-9743
> My opinions are not necessarily those of HP
>
>


-- 
Wojciech Turek

Senior System Architect

High Performance Computing Service
University of Cambridge

Reply via email to