PBS

[Jim Forsythe]

My guess is that you need to let mpi (or whatever you are using) know what nodes to run on.  PBS creates a file with a list of the nodes it has allocated - $PBS_NODEFILE.  But it is up to your application or environment to do something with this file.  You need to pass this to mpi. For mpich, you can edit mpirun.args (in the bin directory) to look like:

if [ "$PBS_NODEFILE" != "" ]
then
  machineFile=$PBS_NODEFILE
fi


Hope this helps.

Jim
  ----- Original Message ----- 
  From: Joey Raheb 
  To: beowulf at beowulf.org 
  Sent: Wednesday, January 17, 2001 10:56 AM
  Subject: PBS


  Hello everyone,
   
  I tried a post on the PBS site and no one seems to want to respond.  I have set up an execution queue on our cluster using PBS.  However, I am having problems getting the jobs to distribute themselves among the different nodes and different CPUS on the nodes.  If I send three jobs at the same time to the queue they all seem to end up on the same node on the same CPU?  Why is that and how do I change that?

  Joey
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://www.scyld.com/pipermail/beowulf/attachments/20010117/8ed62d03/attachment.html