[GE users] Default Architecture .. and How to OverRide It ..

Reuti reuti at staff.uni-marburg.de
Mon Nov 12 14:25:29 GMT 2007


Am 12.11.2007 um 04:38 schrieb Graham Jenkins:

> We're currently running a cluster of lx-24-amd64 execute nodes.   
> We've placed these in three
> separate queues according to allowed max-walltime and memory-size,  
> and we've instructed users to
> submit their jobs with a resource-requirement specification rather  
> than a queue-name so that
> (for instance) large-memory slots can be used if there are no  
> default-memory slots available.

great! Even with this configuration you can still have one queue:

h_rt 3600,[@long_ones=86400],[@ultra_ones=2592000]

if you have appropriate hostgroups.

> But now it seems that (to satisfy some commercial considerations)  
> we'll be installing some
> Solaris nodes.
> So the question is .. how can we decree that a job will go to lx-24- 
> amd64 slot(s) unless a user
> requests otherwise?

You can put this in a default request in $SGE_ROOT/default/common/ 
sge_request like

-l arch=lx24-amd64

The user can still request -l arch=sol-amd64 and replace this  
request. To remove it completely you would need -clear in qsub.

One step more would be to have a shared directory with a similar  
structure for each architecture. This could look like:


and to run "/opt/chemsoft/$ARC/g03/g03" in the jobscript - you always  
get the correct binaries in case you don't care where it runs after  
using -clear.

-- 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