How to use matlab¶
Software

Version

Cluster


matlab

r2020b

Dardel

Getting Started with serial and parallel MATLAB¶
Load the matlab module by:
ml add PDC/22.06
ml matlab/r2020b
In order to use MATLAB you need to have a MATLAB license. KTH has a site license that enable academic users of PDC computers around the world (student, faculty, staff) to use MATLAB. To access the installation of MATLAB on Dardel, please contact PDC support and request access.
Running interactively¶
MATLAB can be run interactively on an allocated node. To book a single node for one hour, type:
salloc N 1 t 1:00:00 A pdc.staff p main
# wait for a node to be reserved
A typical output will look like:
salloc: Granted job allocation 591571
salloc: Waiting for resource configuration
salloc: Nodes nid001015 are ready for job
Node nid001015 is now yours for the next hour, and you can log into it and start MATLAB. On Dardel you can login to the reserved node via the login node:
ssh X nid001015
and then start MATLAB with:
ml add PDC/22.06
ml matlab/r2020b
matlab
In case you do not need a full node with 128 cores, you could request cores in the shared partition. These cores are shared with other users, with the amount of memory provided proportional to the number of cores awarded.:
salloc n 24 t 1:00:00 A pdc.staff p shared
In the following example, a parallel pool of 24 workers is opened and a function, parallel_example (described further below), is called:
>> p=parpool(24)
Starting parallel pool (parpool) using the 'local' profile ... connected to 24 workers.
p =
Pool with properties:
Connected: true
NumWorkers: 24
Cluster: local
AttachedFiles: {}
IdleTimeout: 30 minute(s) (30 minutes remaining)
SpmdEnabled: true
>> parallel_example
t =
2.4711
ans =
2.4711
This is how parallel_example.m might look:
function t = parallel_example
t0 = tic;
parfor idx = 1:24
A(idx) = idx;
pause(2)
end
t = toc(t0)
Running parallel batch jobs¶
You can also submit parallel workflows to the SLURM queueing system. The following job script allocates 16 cores on Dardel and runs one MATLAB program. If the program contains the parpool() command and parfor loops (as demonstrated above), MATLAB will parallelize the computation over different cores of the node.
#!/bin/bash l
# The l above is required to get the full environment with modules
# Set the allocation to be charged for this job
# not required if you have set a default allocation
#SBATCH A snicYYYYXXX
# The name of the script is myjob
#SBATCH J myjob
# The partition
#SBATCH p shared
# The number of cores requested
#SBATCH n 16
# 10 hours wallclock time will be given to this job
#SBATCH t 10:00:00
# Load the Matlab module
ml add PDC/22.06
ml matlab/r2020b
# Run matlab taking your_matlab_program.m as input and show the output in the file
# your_matlab_program.out. The input file must be in the directory where you submit this script.
# This is also where the output will be created.
matlab nosplash nodesktop nodisplay < your_matlab_program.m > your_matlab_program.out
# If you remove the part:
# > your_matlab_program.out
# the output will instead be shown in the default file slurmXXX.out
# where XXX is the job id
Update the SLURM directives to set your project ID and the number of nodes and hours that you wish to allocate, and save the script as run_matlab.sh
. You can then submit the job with:
sbatch run_matlab.sh
It is advisable to run your code with different numbers of workers to determine the ideal number to use.
Running multiple serial batch jobs¶
If you have several serial MATLAB jobs that can be run simultaneously (for example, one MATLAB program with different input parameters), these can be run simultaneously on a booked node (in the main partition) or on booked cores (in the shared partition). A simple bash script for submitting multiple MATLAB jobs on 24 cores in the shared partition of Dardel node can look like
#!/bin/bash l
# The l above is required to get the full environment with modules
# Set the allocation to be charged for this job
# not required if you have set a default allocation
#SBATCH A snicYYYYXXX
# The name of the script is myjob
#SBATCH J myjob
# The partition
#SBATCH p shared
# The number of cores requested
#SBATCH n 24
# 10 hours wallclock time will be given to this job
#SBATCH t 10:00:00
# Load the Matlab module
ml add PDC/22.06
ml matlab/r2020b
# Run matlab for 24 individual programs serial_program_1.m,
# serial_program_2.m ... and print output in files logfile_1, logfile_2, ...
# The input files must be in the directory where you submit this script.
# This is also where the output will be created.
for i in {1..24} ; do
matlab nosplash nodesktop nodisplay < serial_program_${i}.m > logfile_$i &
done
# this wait command must be present since otherwise the job will terminate immediately
wait
Update the SLURM directives to set your project ID and the number of cores and hours that you wish to allocate, and save the script as run_matlab.sh
. You can then submit the job with:
sbatch run_matlab.sh
To learn more¶
To learn more about the MATLAB Parallel Computing Toolbox, check out these resources:
Disclaimer¶
PDC takes no responsibility for the correctness of results produced with the binaries. Always evaluate the binaries against known results for the systems and properties you are investigating before using the binaries for production jobs.