Re: [HTCondor-users] Regarding CPU pinning mechanism in condor

[Ben Cotton <ben.cotton@xxxxxxxxxxxxxxxxxx>]

On Wed, Jan 21, 2015 at 2:59 AM, Raman Sehgal <sc.ramansehgal@xxxxxxxxx> wrote:

> I just wanted to know that once i submit the job and condor allocates some
> slot to it.
> Then whether my job keeps on running on the same slot , which is allocated,
> till the job finished.
> or in between it may switch to some other slot (provided the cluster is
> running smoothly, none of the node goes down during execution).
>
That depends on your configuration. A running job may be preempted
based on the rank of idle jobs or other settable policies (e.g.
keyboard activity, disk utilization). See, e.g,
http://research.cs.wisc.edu/htcondor/manual/v8.2/2_7Priorities_Preemption.html

> or if is it possible, then during switching it may go to some other slots,
> but this case will not provied the efficient use of resouce because the job
> will not be able to make use of data in cpu cache.
>
You can enforce CPU affinity with the ENFORCE_CPU_AFFINITY configuration macro:

A boolean value that defaults to False. When False, the affinity of
jobs and their descendants to a CPU is not enforced. When True,
HTCondor jobs and their descendants maintain their affinity to a CPU.
When True, more fine grained affinities may be specified with
SLOT<N>_CPU_AFFINITY.

However, this is different than the scheduler putting the job on a
different slot, which would require preemption of the job. The
scheduler isn't going to "get bored" and move jobs around except as
specified by your configuration policy.


Thanks,
BC

-- 
Ben Cotton
main: 888.292.5320

Cycle Computing
Better Answers. Faster.

http://www.cyclecomputing.com
twitter: @cyclecomputing