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

Re: [HTCondor-users] Affinity of interactive jobs for a particular node?



OK, I'll check that out. Another option I just looked at is:

(RemoteOwner =?= UNDEFINED) * (DetectedCpus - TotalLoadAvg - 1.0e10*(Offline=?=True))

Which should rank based on the load average relative to detected CPUs. This would be particularly helpful for our partitionable slots. When I run a condor_status and use format to display current values for our machines, it seems to have a reasonable rank.

--
Nathan Smith
Research Systems Engineer
Advanced Computing Center
Oregon Health & Science University

________________________________________
From: HTCondor-users [htcondor-users-bounces@xxxxxxxxxxx] on behalf of Todd Tannenbaum [tannenba@xxxxxxxxxxx]
Sent: Wednesday, August 05, 2015 2:00 PM
To: HTCondor-Users Mail List
Subject: Re: [HTCondor-users] Affinity of interactive jobs for a particular node?

On 8/5/2015 3:42 PM, Nathan Smith wrote:
> Very good, thanks for the tip. I found that in our environment this was set to:
> (RemoteOwner =?= UNDEFINED) * (KFlops - SlotID - 1.0e10*(Offline=?=True))
>
> I also noticed using condor_status that we have highly variable KFlops within our environment, and that the distribution of jobs indeed matches the rank of kflops (assuming the the machine classad matches the requirements of the job).
>
> Any tips for smoothing the KFlops value when used to calculate NEGOTIATOR_POST_JOB_RANK ?
>

You could consider using the ClassAd function quantize() - a list of the
built-in ClassAd functions, including documentation for quantize(), for
HTCondor v8.0 is available at http://goo.gl/iwcwPg

So the value of quantize(KFlops, 1000) will be 1000 if KFlops=50 or 990,
will be 2000 if KFlops=1100 or 1900, etc.  Handy for bucketing values
that are about equal.

hope this helps,
Todd


> --
> Nathan Smith
> Research Systems Engineer
> Advanced Computing Center
> Oregon Health & Science University
>
> ________________________________________
> From: HTCondor-users [htcondor-users-bounces@xxxxxxxxxxx] on behalf of Todd Tannenbaum [tannenba@xxxxxxxxxxx]
> Sent: Tuesday, August 04, 2015 1:44 PM
> To: HTCondor-Users Mail List
> Subject: Re: [HTCondor-users] Affinity of interactive jobs for a particular node?
>
> On 8/3/2015 10:42 AM, Nathan Smith wrote:
>> We have observed that htcondor appears to select a favored machine for interactive jobs. The result is that we'll have multiple interactive jobs assigned to a single machine, even as htcondor is matching non-interactive jobs in a more distributed manor. Is there anything which can be done to avoid clumping interactive jobs on a single node?
>>
>> We're using HTcondor 8.0 branch, and have partitionable slots configured.
>>
>
> Perhaps just changing NEGOTIATOR_POST_JOB_RANK
> (see http://goo.gl/ZHnPhC)  in your condor_config on your central
> manager machine (and doing a condor_reconfig of course) to something
> random like
>
>     NEGOTIATOR_POST_JOB_RANK = random(10000)
>
> ?
>
> There are fancier things done by default in more recent versions of
> HTCondor, but I think the above may do the trick for you in v8.0.x..
>
> regards
> Todd
> _______________________________________________
> HTCondor-users mailing list
> To unsubscribe, send a message to htcondor-users-request@xxxxxxxxxxx with a
> subject: Unsubscribe
> You can also unsubscribe by visiting
> https://lists.cs.wisc.edu/mailman/listinfo/htcondor-users
>
> The archives can be found at:
> https://lists.cs.wisc.edu/archive/htcondor-users/
> _______________________________________________
> HTCondor-users mailing list
> To unsubscribe, send a message to htcondor-users-request@xxxxxxxxxxx with a
> subject: Unsubscribe
> You can also unsubscribe by visiting
> https://lists.cs.wisc.edu/mailman/listinfo/htcondor-users
>
> The archives can be found at:
> https://lists.cs.wisc.edu/archive/htcondor-users/
>


--
Todd Tannenbaum <tannenba@xxxxxxxxxxx> University of Wisconsin-Madison
Center for High Throughput Computing   Department of Computer Sciences
HTCondor Technical Lead                1210 W. Dayton St. Rm #4257
Phone: (608) 263-7132                  Madison, WI 53706-1685
_______________________________________________
HTCondor-users mailing list
To unsubscribe, send a message to htcondor-users-request@xxxxxxxxxxx with a
subject: Unsubscribe
You can also unsubscribe by visiting
https://lists.cs.wisc.edu/mailman/listinfo/htcondor-users

The archives can be found at:
https://lists.cs.wisc.edu/archive/htcondor-users/