On 9/20/2012 8:26 AM, Slimane Amar wrote:
Hi, We have setup a pool of 2 machines : M1(8 cpus) and M2(24 cpus) with a single partitionable slot : MAX_SLOT_TYPES = 1 SLOT_TYPE_1 = 100% NUM_SLOTS_TYPE_1 = 1 SLOT_TYPE_1_PARTITIONABLE = TRUE We want to give a preference to the machine that has the most cpu (M2). The submit file is : rank = Cpus #rank = DetectedCpus request_cpus = 4 We want alls jobs run with 4 cpus on M2 first and then if there is no cpu on M2 run on M1. But with the above configuration, the jobs run on M1 even if there are cpus on M2. What's wrong ? thanks in advance.
Hi Amar -You can achieve the above policy using static slots, but I am not sure you can currently fully achieve the above policy using partitionable slots. So my guess is if you have one job submitted, it will always start on M2, but if you have more than one job, you end up with jobs running on both M1 and M2, right?
This is something we can think about how to improve in the developer series - partitionable slots are still a work in progress in Condor, and we are currently thinking about how to tie up some known shortcomings.
regards, Todd