[Mauiusers] Reservations for GPUs

Denis denismpa at gmail.com
Tue Sep 27 08:07:34 MDT 2011


Hello, Steeve, thank you for your reply!

2011/9/27 Steve Crusan <scrusan at ur.rochester.edu>:
> -----BEGIN PGP SIGNED MESSAGE-----
> Hash: SHA1
>
>
> On Sep 27, 2011, at 9:45 AM, Denis wrote:
>
>> Hello, Burkhard
>>
>> Burkhard Bunk <bunk <at> physik.hu-berlin.de> writes:
>>
>>>
>>> Hi Henrik,
>>>
>>> I had a similar problem recently.
>>> Since "consumable resources" didn't work for me, I used standing
>>> reservations to "split the nodes":
>>> Hope that helps.
>> with this approach do you have any mechanism that can allow the user to figure
>> out which gpu is in use and which one is available when he gets allocated into a
>> node?
>
>
> You can use the PBS_GPUFILE as an indicator to which GPU a user has been assigned.
>
> http://www.clusterresources.com/torquedocs21/3.7schedulinggpus.shtml
I am able to schedule gpus with torque as long as I dont use maui as
scheduler, because it will just not start the jobs when gpus are
requested...
>
> In reality, this is merely a placeholder value, because I'm not sure TORQUE physically denies access to the GPU. You can however >use prolog/epilog scripts to set user permissions, but in my experience people haven't abused or mistakenly used the wrong GPUs.
>
We are using Gromacs with openmm and it takes a paraemter with the
index of the gpu on which it should start running. It feels very
comfortable to use the PBS_GPUFILE as an indexer. The problem is that
we'd like to use fairshare within our scheduler, and torque-sched will
not provide it.

> If you use Moab as your scheduler, Moab can automatically will track GPUs as a consumable resource.
>
Do you know if there is any workaround to get maui+torque scheduling gpus?

I have googled for a while and so far, every result makes tend believe
that it does not work.

>
> ~Steve
>
>
>>
>> Do you still use this fashion for scheduling gpus?
>>
>>>
>>> Regards,
>>> Burkhard Bunk.
>>
>> Thank you in advance,,
>> Denis.
>>
>>
>> --
>> Denis Anjos,
>> www.versatushpc.com.br
>> _______________________________________________
>> mauiusers mailing list
>> mauiusers at supercluster.org
>> http://www.supercluster.org/mailman/listinfo/mauiusers
>
>  ----------------------
>  Steve Crusan
>  System Administrator
>  Center for Research Computing
>  University of Rochester
>  https://www.crc.rochester.edu/
>
>
> -----BEGIN PGP SIGNATURE-----
> Version: GnuPG/MacGPG2 v2.0.17 (Darwin)
> Comment: GPGTools - http://gpgtools.org
>
> iQEcBAEBAgAGBQJOgdX9AAoJENS19LGOpgqK5CYH/jCqef+zdw2DWUe6rHVaAea3
> Zt1d446SeelMDmptdLtOPgGT1LWZc9sdnLZH6Su0nmzi90S+jw5ZFNEATXhtq5oX
> RVMFe8WROFBio3oBeDlZFldPgAmuA6FLXyiUY3x9JYtoFOX1cdmdWgcIqRX5rvWH
> pW7W0603fxCSZdg0Lxgwg9HfbHEQznuSpPgc8AxPkheIKdCn0mt5fWQJv6qFLLEq
> FRTCFbfi+SWpwQy98qJjqpDryQe025ryxH9CxUfaStqqxBVgTIy3DrmfUQkc9stk
> nccfPWdRqqWSGfCf/U8DUGfeGRKd+nFdB+IRzrSw7OoeBuSJrOzyFKugg0/Atrw=
> =xmjP
> -----END PGP SIGNATURE-----
>



-- 
Denis Anjos,
www.versatushpc.com.br


More information about the mauiusers mailing list