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

Re: [Condor-users] GPU machines is "mixed" environment (conditional preemption)

Greg Thain wrote:
 On 10/07/2010 09:19 AM, Carsten Aulbert wrote:
However, now the big question how to address preemption. Essentially, I want to ensure that the machine works as a standard compute nodes with multiple cores (identically setup as others; sans vanilla universe jobs) in the absence
of any idle jobs which have "NeedGpu" set.

I'm not sure I entirely understand your requirements, but I think you can simply
use the Machine RANK to prefer jobs with NeedGpu set.

I'm also a little confused as PREEMPTION_REQUIREMENTS is a condor_negotiator thing but you're talking about slot-level detail. For that you need to use PREEMPT, which is the startd side. You've got the basic formula from here:


...but I think you're wanting to use STARTD_JOB_EXPRS so that you can refer to the attribute of the running job within the slot (MY.NeedsGpu) as well as the candidate job attribute (TARGET.NeedsGpu):


-- Lans Carstensen