[torqueusers] More newbie questions

Gabe Turner gabe at msi.umn.edu
Fri Jan 19 11:07:26 MST 2007


On Fri, Jan 19, 2007 at 12:54:42PM -0500, David Chin wrote:
> OK. I'm trying to wrap my head around Torque. (Maui, too, but that
> will have to wait.) If I could get feedback on my thoughts, that would
> be great. Y'all have been really helpful already.
> 
> So far, this is what I gather.
> * a Cluster (my own term) is made up of one Server and one or more Nodes
> * the Server is the machine which runs pbs_server, and which launches
> and kills jobs
> * Nodes are machines which run pbs_mom, and which run jobs and/or
> provide resources. The Server may or may not be a Node, as well.
> * there are at least two types of Nodes:
>  * cluster
>  * time-shared
> * the Server may have one or more Queues.
> * there are two types of Queues:
>  * Execution Queues, which execute Jobs. Execution Queues have one or
> more Nodes associated with them.
>  * Routing Queues, which insert Jobs into appropriate Queues based on
> available resources, etc.
> * all free Nodes in a Cluster are available to all Queues.

All correct.

> Now, here's a setup that I would like to have. I have 17 compute nodes
> and one head node.  I would like a queue with only the head node on
> for running compilation jobs. This would be in addition to the normal
> execution queues I have to run compute jobs.  Would this require running
> another pbs_server to deal with the compilation queue? 

Nope.  Your single pbs_server can provide multiple queues.  You would
create a 'compile' queue and a 'work' queue.  Now, limiting these queues to
exclusive sets of nodes is where it gets tricky.  For that, refer here:

http://www.clusterresources.com/products/mwm/docs/12.1nodelocation.shtml#open

HTH,

Gabe
-- 
Gabe Turner                                             gabe at msi.umn.edu
UNIX System Administrator,
University of Minnesota
Supercomputing Institute                          http://www.msi.umn.edu


More information about the torqueusers mailing list