[torquedev] Confused about processes again

[Chris Samuel]

On Fri, 26 Feb 2010 06:40:12 pm Simon Toth wrote:

> My problem is that our users also require node-exclusive jobs. Such jobs
> always use the whole node regardless of the amount of resources they
> actually use (for example I/O heavy jobs use this).

Two options that I've used in the past:

1) Make the user who needs the whole node request all the CPUs with (say):

qsub -l nodes=1:ppn=8

2) Define a default pmem or pvmem memory request for all queues and then have 
the user who needs a whole node to submit a job that requests the amount of 
RAM that is available.  For instance when I was at VPAC we had a system that 
was uniformly 4GB per core, so we had a default pmem of just under 4GB.

Our accounting system then scaled their usage (cores*hours) if they had 
exceeded that figure (as there would be cores unavailable to other users, so 
whilst they weren't running code on them they were effectively occupying 
them).

cheers!
Chris
-- 
 Chris Samuel  :  http://www.csamuel.org/  :  Melbourne, VIC

This email may come with a PGP signature as a file. Do not panic.
For more info see: http://en.wikipedia.org/wiki/OpenPGP
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 481 bytes
Desc: This is a digitally signed message part.
Url : http://www.supercluster.org/pipermail/torquedev/attachments/20100228/1e50b486/attachment.bin