La soumission de job se fait principalement via une commande « sbatch script.sh ».

Script de soumission d’un calcul lammps

#!/bin/sh
#SBATCH --job-name=job
#SBATCH --partition=normal            # submission queue (normal or long or bigmem or bigpu or quadgpu)
#SBATCH --time=1-1:00:00            # 1-1 means one day and one hour
#SBATCH --mail-type=ALL    # Type can be BEGIN, END, FAIL, ALL(any statchange).
#SBATCH --output=job_seq-%j.out        # if --error is absent, includes also the errors
#SBATCH --mem=8G    # T-tera, G-giga, M-mega
#SBATCH --nodes=4   # 4 cpus CPU Numbers
#SBATCH --mail-user=votre_mail@domain.precis
echo "-----------------------------------------------------------"
echo "hostname                     =   $(hostname)"
echo "SLURM_JOB_NAME               =   $SLURM_JOB_NAME"
echo "SLURM_SUBMIT_DIR             =   $SLURM_SUBMIT_DIR"
echo "SLURM_JOBID                  =   $SLURM_JOBID"
echo "SLURM_JOB_ID                 =   $SLURM_JOB_ID"
echo "SLURM_NODELIST               =   $SLURM_NODELIST"
echo "SLURM_JOB_NODELIST           =   $SLURM_JOB_NODELIST"
echo "SLURM_TASKS_PER_NODE         =   $SLURM_TASKS_PER_NODE"
echo "SLURM_JOB_CPUS_PER_NODE      =   $SLURM_JOB_CPUS_PER_NODE"
echo "SLURM_TOPOLOGY_ADDR_PATTERN  = $SLURM_TOPOLOGY_ADDR_PATTERN"
echo "SLURM_TOPOLOGY_ADDR          =   $SLURM_TOPOLOGY_ADDR"
echo "SLURM_CPUS_ON_NODE           =   $SLURM_CPUS_ON_NODE"
echo "SLURM_NNODES                 =   $SLURM_NNODES"
echo "SLURM_JOB_NUM_NODES          =   $SLURM_JOB_NUM_NODES"
echo "SLURMD_NODENAME              =   $SLURMD_NODENAME"
echo "SLURM_NTASKS                 =   $SLURM_NTASKS"
echo "SLURM_NPROCS                 =   $SLURM_NPROCS"
echo "SLURM_MEM_PER_NODE           =   $SLURM_MEM_PER_NODE"
echo "SLURM_PRIO_PROCESS           =   $SLURM_PRIO_PROCESS"
echo "-----------------------------------------------------------"

# USER Commands
# Move to /scratch or launch from it
# mv directory
# special commands for openmpi/gcc

module load openmpi/gcc/64/1.10.7

# lammps
module load lammps
./lmp-mpi etc arg agr < file.txt
# end of the USER commands

Lancement de plusieurs tâches dans un même job.

voici une solution si vous souhaitez lancer plusieurs tâches dans un même job,

Admettons que vous souhaitiez lancer 4 commandes en même temps appelés my_cmd1, my_cmd2, my_cmd3 et my_cmd4, alors ajouter ces 4 commandes dans votre script slurm tout en ajoutant un & à la fin de la ligne (ou de la commande). N’oubliez pas d’ajouter la commande wait à la fin de votre script. Sinon votre job terminera avant que les 4 commandes aient eu le temps de s’exécuter. N’oubliez pas de paramétrer le nombre de tâches que vous souhaitez lancer en parallèle (e.g. #SBATCH –ntasks-per-node=4)

Ce qui donne :

#!/bin/sh
#SBATCH --job-name=job
#SBATCH --partition=normal            # submission queue (normal or long or bigmem or bigpu or quadgpu)
#SBATCH --time=1-1:00:00            # 1-1 means one day and one hour
#SBATCH --ntasks-per-node=4            # 1-1 means one day and one hour
# Define all your sbatch parameters you need here
# ...

echo "-----------------------------------------------------------"
echo "hostname                     =   $(hostname)"
echo "SLURM_JOB_NAME               =   $SLURM_JOB_NAME"
# displays all pieces of information that you need
# ...
echo "-----------------------------------------------------------"

./my_cmd1&
./my_cmd2&
./my_cmd3&
./my_cmd4&
wait