[Mauiusers] Parallel Job with more than one compute node doesn't start!
dneedham at cmu.edu
Mon Dec 14 10:39:15 MST 2009
On Sun, 2009-12-13 at 16:28 +0200, Vamvakopoulos Manolis wrote:
> JOBNODEMATCHPOLICY EXACTNODE
This is one parameter which I am not sure I fully understand yet as a
new Maui admin. I know the docs talk about "zero or more of the
following: EXACTNODE or EXACTPROC" in appendix F for this parameter, but
what I have to wonder about is when you end up with that zero case of
neither being specified (the default), or, if possible, both specified.
For example with 8-core nodes, how does it affect the running of jobs
with ENABLEMULTIREQJOBS enabled, or even jobs with a parameter such as
if only EXACTPROC is specified and a node with all 8 cores free is the
only one available.
Only having two of our nodes running Torque at the moment as I do tasks
to prove out the install for the profs in charge of the cluster (such as
can I run a job using all 16 cores using MPI, or submit a job which runs
as a different user without blocking, regardless of the trick I had to
pull), testing the differences has not been high on my priority list
yet, and resources are lacking. But I know as soon as they decide to
deploy, understanding parameters like this will start to become crucial
as they demand answers in exceedingly short time frames, and so...)
I am suspecting that with neither (and perhaps both) specified for the
policy, it comes down to having no constraints, and you get exactly what
you ask for, though if a second job could use the extra cores, it could
be assigned the same node to use those otherwise idle cores. And that
in the case where if you have EXACTPROC specified, if it had to, it
would prefer nodes exactly matching, though it would not block and would
assign it to a node of 8 cores if it had to do so.
More information about the mauiusers