Can't request all the nodes in my cluster

[Roy Dragseth]

On Monday 28 June 2004 16:31, itornes at juno.com wrote:
> 
> I'm running an 8 node cluster using Redhat Linux 9.0.  I set the cluster up using Oscar 3.0 and I'm running OpenPBS 2.3.16 with mpiexec and using the Maui scheduler.  I was not able to patch OpenPBS b/c oscar doesn't install the source files.  mpiexec seems to run just fine without the patch.  
> 
> Each of my nodes has a 3.06 GHz P4 chip with hyper threading supported by the motherboard.  OpenPBS recognizes this and tells me I have 8 nodes each with 2 virtual processors.  
> 
> When I submit a  job via PBS I cannot request more than 8 nodes.  It complains that I have exceeded the available resources.  However if there are no jobs running on the cluster, and I request 8 nodes, it starts the job on 4 of the nodes using 2 processes per node.  This leads me to believe that hyperthreading is working fine.  
> 
> Is my problem a PBS problem, or something to do with mpiexec?  

This is an *pbs artifact, it distinguishes between nodes and cpus.  To utilize all cpus you need to do it like this:
-lnodes=8:ppn=2.

ppn = Processors Per Node.

r.

-- 

  The Computer Center, University of Tromsø, N-9037 TROMSØ Norway.
	      phone:+47 77 64 41 07, fax:+47 77 64 41 00
     Roy Dragseth, High Performance Computing System Administrator
	 Direct call: +47 77 64 62 56. email: royd at cc.uit.no