[torqueusers] Question about virtual processor allocation

[Garrick Staples]

On Wed, Feb 01, 2006 at 08:40:29AM -0500, Jonas_Berlin at harte-hanks.com alleged:
[..snip..]
> When I submit a job I would like to have a VP allocated to each of the 
> partitions of my process and I don't care where they come from, except I 
> would like them to be packed together as much as possible.
> 
> When I submit a job I would like to simply tell it how many CPUs to run 
> the job on and have Torque pick the appropriate nodes.
> 
> If I specify 
> 
> -l nodes=2 on the command line the first job gets allocated to VPs 
> (machine1/0, machine2/0) and the second job get allocated (machine1/1, 
> machine2/1)
> 
> This has the issue that the two jobs get allocated across two machines 
> each causing unnecessary network communication as the jobs run.

Supposedly setting the node_pack server attribute to true will do what
you want.

  qmgr -c 'set server node_pack = T'

[..snip..]
> Any ideas?? Do I have to get a real scheduler like Maui to be able to do 
> this?

Maui will do this when "NODEMATCHPOLICY" is not set in maui.cfg.

-- 
Garrick Staples, Linux/HPCC Administrator
University of Southern California
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 189 bytes
Desc: not available
Url : http://www.supercluster.org/pipermail/torqueusers/attachments/20060201/0751ce24/attachment.bin