[torqueusers] simulating worst case behavior from my cluster. request 8 cpus such that no two are on the same node.
rpnabar at gmail.com
Mon Apr 20 15:13:36 MDT 2009
On Mon, Apr 20, 2009 at 3:53 PM, Victor Gregorio
<vgregorio at penguincomputing.com> wrote:
> 1) Request 8 nodes at 1 processor each:
> #PBS -l nodes=8:ppn=1
Did not work. I got assigned
node18, node18, node18, node18,node18,node18,node07,node07
Is my PBS/Torque mis-configured or is this the right behavior to expect?
> 2) Request each node by hostname and use 1 processor each:
> #PBS -l nodes=node1:ppn=1+node2:ppn=1+node3:ppn=1 ..
I am trying out this version now but it will take a while to have this
exact set of nodes available.
I wish the first one would have worked though! It would have been the
more flexible and less-restrictive way of doing things.On the other
hand since this is only a benchmark job it does not matter much that I
have an elegant way of invoking it.
More information about the torqueusers