MATLAB (matrix laboratory) is a multi-paradigm numerical computing environment and fourth-generation programming language. Developed by MathWorks, MATLAB allows matrix manipulations, plotting of functions and data, implementation of algorithms, creation of user interfaces, and interfacing with programs written in other languages, including C, C++, Java, and Fortran. (taken from http://en.wikipedia.org/wiki/MATLAB, more details there)

http://www.mathworks.de/products/matlab/

We currently offer the following MATLAB versions:

math/MATLAB/2017a
math/MATLAB/2018b
math/MATLAB/2021b

You can query the versions yourself with module av matlab.

Load the most recent version available with:

$ module load math/MATLAB

You can also load a particular version with module load math/MATLAB/<version>, e.g. module load math/MATLAB/2017a

See the vendor documentation: https://www.mathworks.com/help/matlab/

>> exit(0)
Error using exit
Unrecognized function or variable 'Settings'.

settings

The currently available toolboxes are

If you want to prevent MATLAB from spamming your home directory with the java.log.<number> files, you can set the following environment variable:

export MATLAB_LOG_DIR=~/Path/To/Your/Directory

You can also define this environment variable in your .bashrc file.

Using Matlab is possible several ways, which we would like to briefly introduce here.

Starting MATLAB from a Linux system prompt using matlab gives you several start-up options. We only present an overview of the most commonly used option. A complete list can be found in the MATLAB documentation.

Request resources according to your needs, e.g. with:

[joe_user@login]$ srun -n<ntasks> -p<partition> -A<account> -t<time> --mem<amount> --pty --preserve-env $SHELL  

Load the desired Matlab module:

module load math/MATLAB

Now you can run Matlab. Add -nojvm flag to start Matlab without the Java virtual machine, -nosplash prevents Matlab from displaying the Matlab logo.

matlab -nodisplay -nosplash

Check the version info and the available toolboxes for that version:

>> ver

gives:

-------------------------------------------------------------------------------------------
MATLAB Version: 9.5.0.944444 (R2018b)
...
Operating System: Linux 3.10.0-957.5.1.el7.x86_64 #1 SMP Fri Feb 1 14:54:57 UTC 2019 x86_64
Java Version: Java is not enabled
-------------------------------------------------------------------------------------------
MATLAB                                                Version 9.5         (R2018b)
Curve Fitting Toolbox                                 Version 3.5.8       (R2018b)
Econometrics Toolbox                                  Version 5.1         (R2018b)
...

There are several options to compile your Matlab code to stand-alone executables/libraries. Being independent of licenses is one of the major advantages here, of course. But when running compiled code with the Matlab Runtime Envirenment (MRE) on MOGON you have to consider the threading of your code just as well as when you run Matlab itself.

Option	Description	Comment
-?	Display help message	Cannot be used in a deploytool app.
-m	Generate a standalone application.	Equivalent to -W main -T link:exe. Cannot be used in a deploytool app.
-n	Automatically treat numeric inputs as MATLAB doubles.	Cannot be used in a deploytool app.
-o outputfile	Specify name of final output file.	Adds appropriate extension. Cannot be used in a deploytoolapp.
-R option	Specify run-time options for MATLAB Runtime.	Valid only for standalone applications using MATLAB Compiler. option = -nojvm, -nodisplay, '-logfile filename', -startmsg, and -completemsg filename
-v	Verbose; display compilation steps.

MATLAB Help Center, mcc Command Arguments Listed Alphabetically

If your code doesn't need the full multithreading capability, which often is the case, you should compile your code with:

mcc -m my_mfile.m 

This causes the -singleCompThread option to be applied automatically, which limits MATLAB to a single computational thread. This is enabled by default on MOGON.
Using -singleCompThread also ensures that your standalone code will run on a single computational thread, which not only doesn't frustrate the core scheduler and the other users less, but also improves the performance of your code by spending less time scheduling all the threads on a core.

mcc_example.m

function hellomogon
    fprintf('\n======================\n')
    fprintf('===[ Hello MOGON! ]===\n')
    fprintf('======================\n\n')
    N = maxNumCompThreads;
    fprintf('\nNumber of Computational Threads: %g\n\n', N)
end

module load math/MATLAB

mcc -m -R -nodisplay mcc_example.m -o hello_mogon_single

./hello_mogon_single

======================
===[ Hello MOGON! ]===
======================
 
Number of Computational Threads: 1

Generally, Matlab detects the number of physical cores and opens the same amount of threads to make full use of the multithreading implemented in the built-in functions. By calling

mcc -m -R -multiCompThread my_mfile.m

you obtain multithread code. Often this might be wanted, but you have to make sure that you select the appropriate resources for this then - namely, the appropriate CPUs. Since Matlab wants to use everything on a host you'll have to set #SBATCH --cpus-per-task and an appropriate memory reservation.

mcc_example.m

function hellomogon
    fprintf('\n======================\n')
    fprintf('===[ Hello MOGON! ]===\n')
    fprintf('======================\n\n')
    N = maxNumCompThreads;
    fprintf('\nNumber of Computational Threads: %g\n\n', N)
end

module load math/MATLAB

mcc -m -R -multiCompThread -nodisplay mcc_example.m -o hello_mogon_multi

./hello_mogon_multi

======================
===[ Hello MOGON! ]===
======================
 
Number of Computational Threads: 20

We would like to explain with an typical Hello MOGON! example how you can achieve this.

The trivial MATLAB script has only one line:

hello_mogon.m

fprintf('Hello MOGON!\n')

Notice: Save your file with the .m extension, but call it without the .m extension.

A serial Job requires only one core, the Slurm job script therefore would read as follows:

The Job script (serial_matlab_job.slurm) i

serial_matlab_job.slurm

#!/bin/bash
#SBATCH --partition=smp
#SBATCH --account=<YourHPC-Account>
#SBATCH --time=0-00:02:00
#SBATCH --mem-per-cpu=512 #0.5GB
#SBATCH --ntasks=1
#SBATCH --job-name=serial_matlab
#SBATCH --output=%x_%j.out
#SBATCH --error=%x_%j.err
 
module purge
module load math/MATLAB
 
matlab -singleCompThread -nodisplay -nosplash -batch hello_mogon

The start-up options (-singleCompThread, -nodisplay, -nosplash) used are explaind above. To run the script, you can now simple submit it to the batch system with the following command:

sbatch serial_matlab_job.slurm

After the job completes, you can view the output with:

cat matlab-example.out

which should give you

[ ... ]
Hello MOGON!

as Matlab will not be able to fully utilize the node, but considers every FPU equal to a CPU.

The following example is a job that uses Matlab's Parallel Computing Toolbox. You can benefit from the built-in multithreading provided by Matlab's BLAS implementation, if you code makes use of this Toolbox or if you have intense computations. The following is an example of MathWorks and can be found here.

parfor_matrix.m

slurm_cpus = str2double(getenv('SLURM_CPUS_PER_TASK'));
fprintf('Number of available Slurm CPUs is: %g\n', slurm_cpus);
myCluster = parpool('local',slurm_cpus);
fprintf('The Number of MATLAB Workers is: %g\n', myCluster.NumWorkers);
 
tic
n = 200;
A = 500;
a = zeros(n);
parfor i = 1:n
    a(i) = max(abs(eig(rand(A))));
end
fin = toc;
fprintf('With %g CPUs per Task the calculation took %g seconds.\n', slurm_cpus, fin);    

The Slurm job script is kept simple, with one process that has four cores for multithreading:

parallel_matlab_job.m

#!/bin/bash
#SBATCH --account=<YourHPC-Account>
#SBATCH --job-name=parallel_matlab_job
#SBATCH --output=%x_%j.out
#SBATCH --error=%x_%j.err
#SBATCH --partition=smp
#SBATCH --time=0-00:05:00
#SBATCH --mem-per-cpu=4096 #4GB
#SBATCH --ntasks=1
#SBATCH --cpus-per-task=4
 
module purge
module load math/MATLAB/2018b
 
matlab -nodisplay -nosplash -batch parfor_matrix

To run the script, simply submit it to the scheduler with the following command:

sbatch parallel_matlab_job.slurm

After the job has finished, view the output with the following command:

cat parallel_matlab*.out

which should give you the following output:

[ ... ]
With 4 CPUs per Task the calculation took 17.0605 seconds.

For optimal performance, the value of –cpus-per-task must be adjusted. Use the smallest value that gives you a significant performance gain. The matrix multiplication example from above has the following runtimes on MOGON II with different –cpus-per-task:

Testing computationally intensive operations
For operations where the number of floating-point computations performed per element read from or written to memory is high, the memory speed is much less important. In this case the number and speed of the floating-point units is the limiting factor. These operations are said to have high “computational density”.

A good test of computational performance is a matrix-matrix multiply. For multiplying two $N \times N$ matrices, the total number of floating-point calculations is $$ FLOPS(N) = 2N^3 - N^2 $$

Two input matrices are read and one resulting matrix is written, for a total of $3N^2$ elements read or written. This gives a computational density of $(2N - 1)/3$ FLOP/element. Contrast this with plus as used above, which has a computational density of $1/2$ FLOP/element.

MATLAB Help Center, Measuring GPU Performance

gpu_perf.m

gpu = gpuDevice();
fprintf('Using an %s GPU\n', gpu.Name)
sizeOfDouble = 8;
sizes = power(2, 12:2:26);
N = sqrt(sizes);
mmTimesHost = inf(size(sizes));
mmTimesGPU = inf(size(sizes));
for ii=1:numel(sizes)
    % First do it on the host
    A = rand( N(ii), N(ii) );
    B = rand( N(ii), N(ii) );
    mmTimesHost(ii) = timeit(@() A*B);
    % Now on the GPU
    A = gpuArray(A);
    B = gpuArray(B);
    mmTimesGPU(ii) = gputimeit(@() A*B);
end

gpu_matlab_job.slurm

#!/bin/bash
#SBATCH --account=<YourHPC-Account>
#SBATCH --job-name=matlab_gpu_job
#SBATCH --output=%x_%j.out
#SBATCH --error=%x_%j.err
#SBATCH --partition=m2_gpu
#SBATCH --gres=gpu:1
#SBATCH --time=0-00:35:00
#SBATCH --mem-per-cpu=4096 #4GB
#SBATCH --ntasks=1
#SBATCH --cpus-per-task=4
 
module purge
module load math/MATLAB/2018b
 
matlab -nodisplay -nosplash -batch gpu_perf

cat gpu_job_*.out

[ ... ]
Achieved peak calculation rates of 72.9 GFLOPS on Host and 390.7 GFLOPS on GPU

In the introduction on compiling m-Files in this article, you have probably observed that even the simple Hello MOGON!-Funcion took its moment to compile into a stand-alone c application. With larger scripts, of course, compile time will not improve, which is why you should always compile m-File in a job on the local scratch directory.

The following is a short, simple example demonstrating how to create such a Job that compiles m-files on the localsratch directory.

Start by creating a new folder and change to it, e.g.:

mkdir matlabEx && cd matlabEx

You can simple delete the entire folder after working through this How-To.<br>

We will re-use the familiar Hello MOGON!-script from the previous examples:

hello_mogon.m

function hellomogon
    fprintf('\n======================\n')
    fprintf('===[ Hello MOGON! ]===\n')
    fprintf('======================\n\n')
end

The job script is a little bit more complex. In the first lines, the usual and necessary job options are given. With the line #SBATCH –signal=B:SIGUSR2@60 we send the signal SIGUSE2 to the application one minute before the job reaches its walltime. We then define variables for the current working directory and an ouput folder where we want to save our resulting files. After loading the proper MATLAB module, we define the cleanup function, which simply copies all files from the localscratch directory to the output directory, either after the job completed successfully or one minute before the job reaches its time limit. The subsequent lines only create folders or changes into them. The actual compilation takes place in the line mcc -m … already on the localscratch directory of the compute node. Of course, you can provide the compiler with all the options that are given above or the ones you can find in the MATLAB documentation.

compileFile.sh

#!/bin/bash
 
#SBATCH --account=<YourHPC-Account>
#SBATCH --partition=smp
#SBATCH --constraint=skylake
#SBATCH --ntasks=1
#SBATCH --cpus-per-task=6
#SBATCH --time=0-00:05:00
#SBATCH --mem=2048
#SBATCH --signal=B:SIGUSR2@60 # e.g. signal 1min before job will end - time is in seconds
 
# Store working directory to be safe
# Define Input- and Outputfile
SAVEDPWD=$(pwd)
FILENAME=$1
OUTNAME="hello_mogon_compJob"
 
module load math/MATLAB
 
# We define a bash function to do the cleaning when the signal is caught
cleanup(){
    # Note: The following only works on single with output on the node,
    #       where the jobscript is running.
    #       For multinode output, you can use the 'sgather' command or
    #       get in touch with us, if the case is more complex.
    cp -R /localscratch/${SLURM_JOB_ID}/* ${SAVEDPWD}/output &
    wait
    exit 0
}
# Register the cleanup function when SIGUSR2 is sent,
# one minute before the job gets killed
trap 'cleanup' SIGUSR2
 
# Create the output directory
mkdir ${SAVEDPWD}/output
 
# Copy input file
cp ${SAVEDPWD}/$FILENAME /localscratch/${SLURM_JOB_ID}
 
# Go to jobdir and start the program
cd /localscratch/${SLURM_JOB_ID}
 
mcc -m -R -multiCompThread -nodisplay $FILENAME -o $OUTNAME
 
# Call the cleanup function when everything went fine
cleanup

Submit the job as usual with the following command:

sbatch compileFile.sh hello_mogon.m

Out cleanup function copied all results to the output directory, after the stand-alone application has been created, run it as follows:

./output/hello_mogon_compJob

The output on the command line should look similar like this:

======================
===[ Hello MOGON! ]===
======================