4-Multi-process + multi-threaded submission

The following takes the sbatch submission method as an example
hybrid -pro-thr command during the calculation process , apply for 2 nodes for resources, each node runs a process, and each process runs 36 cores, then use sbatch to submit the command ( the script name is hybrid-pro-thr.sh ) :

#!/bin/bash
#SBATCH -J hybrid-pro-thr
#SBATCH -o job.%j.out
#SBATCH -p q_cn
#SBATCH -N 2
#SBATCH --ntasks-per-node=1
#SBATCH -c 36
module load anaconda3/4.8.2 #module loads the required software
srun -n 2 ./hybrid-pro-thr

After editing the script, you can submit it directly to the computing node to run

sbatch hybrid-pro-thr.sh

Parameters involved in the example :

-J hybrid-pro- thr # hybrid-pro- thr is the name of the submitted job, custom
-p q_cn #The specified partition for job submission is the q_cn queue;
-o job.% j.out #The output of the script execution will be saved in the job.% j.out file, where %j represents the job number;
--n tasks - per-node = 1 #Run a task ( process ) on each node
-c 36 # use 36 cores per process
-N 2 # 2 nodes

Common submission parameters:

--help # Display help information;
-D, --chdir = # Specify the working directory;
--get-user-env # Get the current environment variables;
--gres = #Required parameters when using gpu card, such as applying for 1 gpu --gres = gpu:1
-J, --job-name= # Specify the job name of the job;
--mail-type= # When the specified state occurs, send email notification, the valid types are (NONE, BEGIN, END, FAIL, REQUEUE, ALL);
--mail-user= #Send to the specified mailbox ;
-n, -- ntasks = #By default, one task is one core ;
-c, --cpus -per-task=< ncpus > # The number of cores required by each task, the default is 1;
-- ntasks -per-node=< ntasks > # The number of tasks per node, the priority of the -- ntasks parameter is higher than this parameter, if the -- ntasks parameter is used , it will become the most running one per node number of tasks;
-o, --output= # Output file to which the output from the job script will be output;
-p, --partition=< partition_names > # Submit the job to the corresponding partition;
-t, --time= # The maximum time allowed for the job to run, the current cluster default time is 7 days
-w, -- nodelist = # Specify the node to apply for;
-x, --exclude= # Exclude the specified node;
- -mem-per- cpu = #The memory size allocated by each core can use the suffix [K|M|G|T] to specify different units