[torqueusers] How to achieve round robin queueing

Stephen J. Scheck ss543 at cornell.edu
Tue Jun 28 13:28:38 MDT 2005


I have a cluster of 9 nodes and head node which is also configured to 
run jobs. All nodes are listed in the server nodes file. Each has 2 
processors. I am running the default torque FIFO scheduler. I am trying 
to achieve the following queueing policy: I want no more than one job 
per processor to run on each node, and when jobs are submitted, I would 
like the job to run first on a node which has no jobs currently running. 
So each time a job is submitted, it should choose a node without any job 
running, then when all nodes have one job, it should schedule the job on 
the 2nd processor of a node, etc.

To make this work, I added the following to the server configuration:

set server node_pack = False

Unfortunately, what is happening is that all jobs are being scheduled on 
one node, usually the first listed in the nodes file, but sometimes on 
other nodes.

What am I missing?



More information about the torqueusers mailing list