launch matlab with mpi on multiple nodes, missing output files

Hello , I am running matlab program with a batch script which lunches multiple nodes. Within each nodes, there are several independent jobs that are paralleled by parfor loop. Inside the parfor, there is a for loop for saving files.

The code that I am running can be simplified to the following examples.

The matlab code looks like:

-----------------------------------------------------------

 my_par=parcluster('local');
 my_parallel_job=parpool(my_par,16);             

 parfor n_loop=1:16

    for n_loop_2=1:5

    fileid=fopen([my_par.Host num2str(n_loop) '_' ...
    num2str(n_loop_2)'.txt'],'w');

    fprintf(fileid,'%d \n',n_loop_2);

    fclose(fileid);

    end                 
end

 delete(my_parallel_job);    

 exit;

-------------------------------------------------------------

Here is my batch script:

---------------------------------------------------

#!/bin/bash

#SBATCH --ntasks-per-node=1

#SBATCH --partition=cfel

#SBATCH -t 00:10:00

#SBATCH --nodes=2

#SBATCH --job-name name_lalala

#SBATCH --output %j-%N.out

#SBATCH --error %j-%N.err

. /etc/profile.d/modules.sh module load mpi/mpich-3.2-x86_64 module load matlab

mpirun -np 2 matlab -nodisplay -nosplash -r "my_matlab_file_name"

-------------------------------------------------------------

everything works properly when my node number in the batch is 1 (total file number=16*5). But when the nodes are more than 1, some output files are missing.

Could anyone please have any suggestion on what could be happening?

many thanks Lu

Answer 1

[lu wang]

answer to myself. Use srun and tasks-per-node instead of mpirun.

problem solved

[venkat K]

Hi lu wang....Could you please check this question [https://ch.mathworks.com/matlabcentral/answers/451144-error-unexpected-matlab-operator-on-cluster] and tell me about the error and how to resolve the error.