Re: [HTCondor-users] Condor + Xeon PHI

[Lans Carstensen <lans.carstensen@xxxxxxxxxxxxxx>]

Greetings!  On Fri, Nov 9, 2012 at 9:38 AM, Tiago Macarios
<tiagomacarios@xxxxxxxxx> wrote:
> Hi Lans,
>
> It looks like that it is "easier" to use than GPUs and if you are using the
> intel compiler it can sometimes be only some extra calls.
> http://goparallel.sourceforge.net/wp-content/uploads/2012/10/SF12_SFTS001_101.pdf

That's where code and workflow specifics matter.  You'll need to dig
in on what the scheduling resource is you're trying to provide.  Is it
acceptable for "accelerated" code to run on a combination of Phi and
non-Phi systems?  That's the difference between Rank and Requirements,
for instance.  The good folks at UW and the various vendors like Red
Hat and Cycle Computing can help you out with workflow specifics.

Regardless of the specifics, In the end state you'll end up adding
resource metadata of some sort to a specialty slot for use in
scheduling, you'll end up with a set of management scripts related to
the hardware on the box, and you'll end up with job submission
templates that match the workflow.

> Also I remember some discussion about multiple GPUs not working on Condor
> when using windows. Would that be a problem with Phi? Not sure it is
> "recognized" as a GPU...

I have no direct knowledge around Condor or Phi cards on Windows, so
I'll pass on that one.

> Thanks,
>
>
> On Fri, Nov 9, 2012 at 2:32 PM, Lans Carstensen
> <lans.carstensen@xxxxxxxxxxxxxx> wrote:
>>
>> On Fri, Nov 9, 2012 at 9:18 AM, Tiago Macarios <tiagomacarios@xxxxxxxxx>
>> wrote:
>> > Hi,
>> >
>> > Have anyone here tried using Condor and Xeon Phi Coprocessor? Would it
>> > work?
>>
>> Short answer is "yes".
>>
>> The slightly longer answer is that they have completely different
>> roles from each other, and will require integration.  Just like GPU's
>> or any other specialty compute hardware - you model and expose the
>> host resources as metadata for your scheduler to utilize and reserve
>> special slots around.  The support for utilizing the specific hardware
>> has to be written into job wrappers and applications, and infers a
>> workflow and specific integration/configuration that you'll have to
>> develop.  You probably want to start with the GPU wiki page:
>>
>> https://condor-wiki.cs.wisc.edu/index.cgi/wiki?p=HowToManageGpus
>>
>> -- Lans Carstensen
>> _______________________________________________
>> 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/