[HTCondor-users] setting memory available to cores

[Hugh Jennings <hugh@xxxxxxxxxxx>]

Hi all,

I am hoping there is a way to assign memory to slots rather than be limited by evenly distributing the RAM between all cores.

The processes we run require a GPU and need as much as 30GB.  Our current systems only have 1 GPU but have up to 8 cores available.  When I start Condor with the default config I see

Name               OpSys      Arch   State     Activity LoadAv Mem   ActvtyTime

slot1@xxxxxxxxxxxx LINUX      X86_64 Unclaimed Idle      0.080 4038  0+00:14:27

slot2@xxxxxxxxxxxx LINUX      X86_64 Unclaimed Idle      0.000 4038  0+00:14:48

slot3@xxxxxxxxxxxx LINUX      X86_64 Unclaimed Idle      0.000 4038  0+00:14:49

slot4@xxxxxxxxxxxx LINUX      X86_64 Unclaimed Idle      0.000 4038  0+00:14:50

slot5@xxxxxxxxxxxx LINUX      X86_64 Unclaimed Idle      0.000 4038  0+00:10:08

slot6@xxxxxxxxxxxx LINUX      X86_64 Unclaimed Idle      0.000 4038  0+00:10:09

slot7@xxxxxxxxxxxx LINUX      X86_64 Unclaimed Idle      0.000 4038  0+00:10:10

slot8@xxxxxxxxxxxx LINUX      X86_64 Unclaimed Idle      0.000 4038  0+00:14:46

                     Total Owner Claimed Unclaimed Matched Preempting Backfill

        X86_64/LINUX     8     0       0         8       0          0        0

               Total     8     0       0         8       0          0        0

I would like to either: 

1. Assign most if not all memory to a single slot
   
2. or remove the other slots from the available pool

I have added the following config:

## GPU Config stuff

SLOT1_HAS_GPU=TRUE

SLOT1_GPU_DEV=0

SLOT2_HAS_GPU=FALSE

SLOT3_HAS_GPU=FALSE

SLOT4_HAS_GPU=FALSE

SLOT5_HAS_GPU=FALSE

SLOT6_HAS_GPU=FALSE

SLOT7_HAS_GPU=FALSE

SLOT8_HAS_GPU=FALSE

STARTD_ATTRS=HAS_GPU,GPU_DEV

to the condor_config.local.

Eventually I want to take proper advantage of HTCondor and have a pool of our machines but I am having problems getting the authorization/authentication working.

I am not a trained admin so I would appreciate instructions or advice that is as explicit as possible.

Regards,

Hugh