Skip to content

Changes

Page history
Mise à jour de Pack arrays of sequential jobs authored by Awa Sepou Ngailo's avatar Awa Sepou Ngailo
Hide whitespace changes
Inline Side-by-side
Running-the-jobs/Pack-arrays-of-sequential-jobs.md
View page @ b9da34ad
## Problem ## Problem
When submitting arrays of jobs, you'll probably want to "pack" them. That is, you want a node to run as many jobs as possible et probably want to avoid sharing your node with another user. When submitting arrays of jobs, you'll probably want to "pack" them. That is, you want a node to run as many jobs as possible et probably want to avoid sharing your node with another user.
Note that this only make sense on sequential partition, and that only those partition support that feature. Note that this only make sense on sequential partition, and that only those partition support that feature.
## How to ## How to
If one single jobs require 4 core, you can specify that in your script with: If one single jobs require 4 core, you can specify that in your script with:
``` ```
#!/usr/bin/env bash #!/usr/bin/env bash
#SBATCH --cpus-per-task=4 #SBATCH --cpus-per-task=4
#SBATCH --partition=seq-short #SBATCH --partition=seq
#SBATCH --time=1:30:00 #SBATCH --time=1:30:00
... ...
<myexe> data_$SLURM_ARRAY_TASK_ID <myexe> data_$SLURM_ARRAY_TASK_ID
``` ```
Then, to launch 20 tasks (that's 5 job per node on a 20 core node): Then, to launch 20 tasks (that's 5 job per node on a 20 core node):
``` ```
$sbatch --array=1-20 --exclusive=user ./cpu-per-task.slurm $sbatch --array=1-20 --exclusive=user ./cpu-per-task.slurm
``` ```
The option '--exclusive=user' indicate that a single node can run more than one job as long as these jobs belong to the same user. The option '--exclusive=user' indicate that a single node can run more than one job as long as these jobs belong to the same user.
### Full example ### Full example
Running the script [packed_array.slurm](https://gitlab.oca.eu/DSI/HPC/-/blob/master/SLURM/sequential/packed_array.slurm) will illustrate the feature: Running the script [packed_array.slurm](https://gitlab.oca.eu/DSI/HPC/-/blob/master/SLURM/sequential/packed_array.slurm) will illustrate the feature:
``` ```
$ sbatch -p seq-short --array=0-20 packed_array.slurm $ sbatch -p seq --array=0-20 packed_array.slurm
Submitted batch job 18985606 Submitted batch job 18985606
$ grep task slurm-18985606_*.out $ grep task slurm-18985606_*.out
slurm-18985606_0.out:Doing task 00000 on node p081.cluster.local slurm-18985606_0.out:Doing task 00000 on node p081.cluster.local
slurm-18985606_10.out:Doing task 00010 on node p081.cluster.local slurm-18985606_10.out:Doing task 00010 on node p081.cluster.local
slurm-18985606_11.out:Doing task 00011 on node p081.cluster.local slurm-18985606_11.out:Doing task 00011 on node p081.cluster.local
slurm-18985606_12.out:Doing task 00012 on node p081.cluster.local slurm-18985606_12.out:Doing task 00012 on node p081.cluster.local
slurm-18985606_13.out:Doing task 00013 on node p081.cluster.local slurm-18985606_13.out:Doing task 00013 on node p081.cluster.local
slurm-18985606_14.out:Doing task 00014 on node p081.cluster.local slurm-18985606_14.out:Doing task 00014 on node p081.cluster.local
slurm-18985606_15.out:Doing task 00015 on node p081.cluster.local slurm-18985606_15.out:Doing task 00015 on node p081.cluster.local
slurm-18985606_16.out:Doing task 00016 on node p081.cluster.local slurm-18985606_16.out:Doing task 00016 on node p081.cluster.local
slurm-18985606_17.out:Doing task 00017 on node p081.cluster.local slurm-18985606_17.out:Doing task 00017 on node p081.cluster.local
slurm-18985606_18.out:Doing task 00018 on node p081.cluster.local slurm-18985606_18.out:Doing task 00018 on node p081.cluster.local
slurm-18985606_19.out:Doing task 00019 on node p081.cluster.local slurm-18985606_19.out:Doing task 00019 on node p081.cluster.local
slurm-18985606_1.out:Doing task 00001 on node p081.cluster.local slurm-18985606_1.out:Doing task 00001 on node p081.cluster.local
slurm-18985606_20.out:Doing task 00020 on node p086.cluster.local slurm-18985606_20.out:Doing task 00020 on node p086.cluster.local
slurm-18985606_21.out:Doing task 00021 on node p086.cluster.local slurm-18985606_21.out:Doing task 00021 on node p086.cluster.local
slurm-18985606_22.out:Doing task 00022 on node p086.cluster.local slurm-18985606_22.out:Doing task 00022 on node p086.cluster.local
slurm-18985606_23.out:Doing task 00023 on node p086.cluster.local slurm-18985606_23.out:Doing task 00023 on node p086.cluster.local
slurm-18985606_24.out:Doing task 00024 on node p086.cluster.local slurm-18985606_24.out:Doing task 00024 on node p086.cluster.local
slurm-18985606_25.out:Doing task 00025 on node p086.cluster.local slurm-18985606_25.out:Doing task 00025 on node p086.cluster.local
slurm-18985606_2.out:Doing task 00002 on node p081.cluster.local slurm-18985606_2.out:Doing task 00002 on node p081.cluster.local
slurm-18985606_3.out:Doing task 00003 on node p081.cluster.local slurm-18985606_3.out:Doing task 00003 on node p081.cluster.local
slurm-18985606_4.out:Doing task 00004 on node p081.cluster.local slurm-18985606_4.out:Doing task 00004 on node p081.cluster.local
slurm-18985606_5.out:Doing task 00005 on node p081.cluster.local slurm-18985606_5.out:Doing task 00005 on node p081.cluster.local
slurm-18985606_6.out:Doing task 00006 on node p081.cluster.local slurm-18985606_6.out:Doing task 00006 on node p081.cluster.local
slurm-18985606_7.out:Doing task 00007 on node p081.cluster.local slurm-18985606_7.out:Doing task 00007 on node p081.cluster.local
slurm-18985606_8.out:Doing task 00008 on node p081.cluster.local slurm-18985606_8.out:Doing task 00008 on node p081.cluster.local
slurm-18985606_9.out:Doing task 00009 on node p081.cluster.local slurm-18985606_9.out:Doing task 00009 on node p081.cluster.local
16:47:56 [alainm@castor sequential]# 16:47:56 [alainm@castor sequential]#
``` ```
\ No newline at end of file