On Mon, Feb 13, 2017 at 03:52:20PM +0000, Mark Dixon wrote: > Hi, > > I've been playing with allocating GPUs using gridengine and am wondering if > I'm trying to make it too complicated. > > We have some 24 core, 128G RAM machines, each with two K80 GPU cards in > them. I have a little client/server program that allocates named cards to > jobs (via a starter method and the handy job gid). > > What's left is the most important question: how do users request these > resources? > > I'm worried that, if I ask them to specify all the job's resources, a > mouthful like "-pe smp 12 -l h_rt=1:0:0,h_vmem=64G,k80=1", just to get one > card, could all too easily result in a K80 sitting idle if the numbers are > tweaked a little. > > Instead, I'm wondering if I should encourage users to request a number of > cards and then we allocate a proportion of cpu and memory based on that (via > per-slot complex, JSV and starter method). > > Is that too simplistic, or would it be a welcome relief? What does the qsub > command line look like at other sites for requesting GPUs?
IIRC Univa Grid Engine has a mechanism for job templates which allow the definition of standard varieties of job. One could implement a similar mechanism via the jsv and contexts. qsub -ac Template=GPU -l gpu=1 script Your jsv could spot that a Template had been requested and fill in sensible defaults based on other requests. If no template is requested users have access to the full power of this fully operational caommand line. I'm not sure what purpose contexts originally served in Grid engine but AFAICT their only practical use is to allow users to pass info to JSVs without affecting anything important. You can do almost anything in Grid Engine with a combination of JSVs and carefully designed started_methods. The hard question is whether you should. William
signature.asc
Description: Digital signature
_______________________________________________ users mailing list users@gridengine.org https://gridengine.org/mailman/listinfo/users
