[Moabusers] Torque/Moab node allocation issue

[VS Ang]

Hello,

I am looking for some help with Torque/Moab configuration. We are using Torque and Moab on a cluster with nodes having 4 cores each. We are trying to schedule jobs such that the nodes run multiple jobs in shared mode. However, the scheduler is not allocating nodes based on the processors. We specified the following policies in the moab.cfg file:

NODEACCESSPOLICY            SHARED
JOBNODEMATCHPOLICY        EXACTNODE

The Torque queue is configured as below:

create queue batch
set queue batch queue_type = Execution
set queue batch resources_max.nodect = 132
set queue batch resources_max.nodes = 132:ppn=4
set queue batch resources_default.nodect = 1
set queue batch resources_default.nodes = 1:ppn=1
set queue batch resources_default.walltime = 01:00:00
set queue batch enabled = True
set queue batch started = True

Also I configured:
set server node_pack = True

The "server_priv/nodes" file has node entries as below:
blade1 np=4
blade2 np=4
blade3 np=4
blade4 np=4
blade5 np=4
..

However, no matter how the job is submitted, users always get 4 different nodes, as opposed to a single node with 4 CPUs.

>msub -I -l nodes=4
>cat $PBS_NODEFILE
blade1
blade2
blade3
blade4

>msub -l nodes=4,taskdistpolicy=pack

Also, another odd behavior is, if I specify:
>msub -l ncpus=4

I am actually getting all nodes in the cluster, as opposed to 4.

So, what is wrong here? Am I missing any other policy parameters in Torque or Moab?

Thanks in advance!

Srihari

-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://www.supercluster.org/pipermail/moabusers/attachments/20071202/4b6eb8b7/attachment.html