Replies: 2 comments 27 replies
-
Hello You can limit the number of concurrent apps using the min/max procs per app parameters. If you still want only one processor to run with several assigned, then use a singleRankPartitioner in the Paritioning block to force it back to one processor Guillaume |
Beta Was this translation helpful? Give feedback.
-
Also, make sure you use the batch-reset mode in combination with your min/max procs per app settings. Here is a short description on the gains: The number of kernels would not influence your memory consumption too much unless you take it to the extreme. |
Beta Was this translation helpful? Give feedback.
-
I am using MOOSE's polynomial chaos for training on a cluster, calculating 256 deterministic samples. Using more kernels will keep initializing the equation, while using fewer kernels will exceed memory or cause the following errors:
srun:First task exited 60s ago
srun:StepId=128229.0 task 3:running
srun :StepId=128229.0 tasks 0-2:exited
srun:launch/slurm:step signal:Terminating StepId=128229.0
srun :srun: Job step aborted: Waiting up to 32 seconds for job step to finish.130131
srun: error: ion171:task 3:Killed
when calculating the second step. Running the main input file alone can run it. Is there a way to abandon multiple apps and execute it in serial order? Or how to group the training sampler and specify the number of kernels for a single example
Beta Was this translation helpful? Give feedback.
All reactions