[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [Condor-users] Condor Parallel universe and sending jobs on same physical nodes.

Hi Manan,

Using the following on your execute nodes would mean that a single job would only ever run on the same physical machine:

ParallelSchedulingGroup     = "$(HOSTNAME)"
STARTD_EXPRS                = $(STARTD_EXPRS), DedicatedScheduler

You can of course bunch machines together by putting groups of them in different ParallelSchedulingGroups.


Manan Chopra wrote:
Dear all,
We have Condor set up for parallel universe of our cluster. Our cluster has nodes which have two cpus and each cpu is dual core, so
we have four cores on each node.  We use condor to run parallel jobs on
our cluster. Usually we run jobs requiring 8 to 12 cores. When a job,
for e.g. requiring 8 cores is submitted to the cluster it tends to select 8 cores to run the job from many different nodes, we would ideally like to run a 8 core job on two nodes.
Is there any way in condor to enforce this criteria for forcing parallel
jobs to run on cores on a same physical node.

Thanks in advance. I will very much appreciate any response in this