Matlab

Research Computing (RC) provides a large suite of software on RC resources. In this tutorial we will learn how to run Matlab on these resources. The tutorial assumes you are familiar with Matlab and basic Linux terminal commands.

There are two basic ways to run Matlab (or many other kinds of software) on RC resources. The first is through an interactive job, and the second is through a batch job. An interactive job allows one to work in real-time with Matlab. Two reasons you may want to do this would be if you are actively debugging your code, or if you would like to use the GUI (in this instance, the Matlab Desktop). However, there might be other reasons you would like to work interactively with Matlab.

The second way to run Matlab on RC resources is through a batch job. This allows the job to run in the background when resources become available. You may choose to use this method if you have a large job that may wait in the queue for awhile, or if you are not debugging or in need of a GUI. Both ways to work with Matlab are below.

Running Matlab Interactive Jobs

Running Matlab interactive jobs on RC resources is both a simple and easy task to accomplish. In this section we will learn 2 ways to launch Matlab as an interactive job:

  1. MATLAB Interactive Application (CURC OnDemand):

    You can access a full Matlab GUI in your browser with CURC OnDemand. CURC OnDemand is a browser-based, integrated, single access point for HPC resources. It includes access to interactive apps such as Matlab, a virtual desktop, and more.

    You can find instructions to start an interactive Matlab session in our OnDemand documentation.

  2. Interactive SLURM job:

    For more information on launching interactive jobs check out our interactive jobs tutorial

    Begin by launching an interactive job by loading slurm/alpine into your environment and running the sinteractive command.

    module load slurm/alpine
sinteractive

    From here you will load the Matlab module into your environment.

    module load matlab

    Lastly we will run Matlab from the terminal.

    matlab -nosplash

    By default Matlab will load an interactive terminal session. If you would like to access the Matlab GUI then simply run Matlab with X11 forwarding enabled.

    To find out how you enable X11 forwarding in your terminal session, check out our X11 forwarding tutorial.

Running Matlab Batch Jobs

Here, we will learn how to run a Matlab script in a non-interactive batch job. For more general information on batch job scripts on Alpine, please see our tutorial on batch jobs

Let’s begin by constructing a small Matlab script that prints ‘hello world’ to the user. The Matlab script we will use for the purposes of this tutorial is called hello_world.m and contains only one line, the Matlab command:

fprintf(‘Hello world\n’)

Which simply prints “Hello world” when called.

Next, we will construct a batch script that will enable us to run this job. The batch script organizes the variety of flags slurm needs to run a job and specifies the software commands we want to execute. One advantage of batch scripts is that they are easily reusable and adaptable for other similar tasks.

We will run this job using a bash script titled: slurm_hello.sh, which contains the following lines:

#!/bin/bash

#SBATCH --nodes=1
#SBATCH --time=0:01:00
#SBATCH --partition=atesting
#SBATCH --qos=testing
#SBATCH --ntasks=1
#SBATCH --job-name=Matlab_Hello_World
#SBATCH --output=Matlab_Hello_World.out

module purge

module load matlab

matlab -nodesktop -nodisplay -r ‘clear; hello_world;’

This file has a few basic parts:

  1. The first line specifies that it is a bash shell script, and ensures the rest of the lines will be interpreted in the correct shell.
  2. The lines beginning with #SBATCH specify the Slurm parameters that will be used for this job. These lines are viewed as comments by bash, but will be read by Slurm. Of particular note is the --output parameter which specifies the file where stderr and stdout (including the output from our Matlab script) will be written. For a description of the Slurm parameters, please see the general Slurm documentation here
  3. The lines beginning with module purge remove any unneeded software and ensure that the appropriate Matlab module is loaded on the compute node.
  4. The final line calls Matlab and instructs it to run our script. This entire line includes commands that are specific to Matlab; the nodesktop and nodisplay flags ensure that the Matlab Desktop will not open, and the r flag will run the script hello_world. The clear command forces Matlab to clear any existing variables, and is simply included as good coding practice.

You have now completed your batch script. After saving the script and exiting your text editor, run the job as follows:

sbatch slurm_hello.sh

Once the job has run, the output of the Matlab script, “Hello world” will be shown in Matlab_Hello_World.out.

Parallel Matlab on Alpine

To fully utilize the multi-core capabilities of Alpine to speed up jobs, most code must first be parallelized. Matlab has many built in tools to accomplish this task. In this tutorial we will parallelize our “Hello World” program.

Let’s begin with the Matlab script we created above called hello_world.m. First we will modify the fprintf line so that it includes a variable ‘i’ that will print out the iteration of the parallel loop.

fprintf(“Hello World from process %i”, i)

Next, we need to encapsulate the print statement in a parallel ‘for’ loop. Matlab uses the construct parfor to separate the task into multiple threads. In order to utilize the parfor command one must ensure that the Parallel Computing Toolbox is available as part of the Matlab software package. CURC has this available and thus no additional action is required on your part if you are utilizing CURC resources.

The order of runtime in the loop is not guaranteed, so the output may not be in sequential order. The loop is formatted as such:

parfor(i = initial_Value:final_Value, maximum_amount_of_threads)

For example, let’s use parfor to implement an 5-iteration loop with a maximum of 4 processors in our script:

parfor(i = 1:5, 4)
        fprintf("Hello, World from process %i", i)
end

Now all we have left to do is modify our batch script to specify that we want to run 4 tasks on the node (we can use up to 64 cores on each ‘amilan’ node on Alpine). We can also change the name of the job and the output file if we choose.

#!/bin/bash

#SBATCH --nodes=1
#SBATCH --time=0:01:00
#SBATCH --partition=atesting
#SBATCH --qos=testing
#SBATCH --ntasks=4
#SBATCH --job-name=Matlab_Parallel_Hello
#SBATCH --output=Matlab_Parallel_Hello.out

module purge

module load matlab

matlab -nodesktop -nodisplay -r 'clear; hello_world;'

Now we run the job using the sbatch command shown above, and our output in Matlab_Parallel_Hello.out will be as follows (the process order may be different in your output):

Hello World from process 4
Hello World from process 1
Hello World from process 2
Hello World from process 3

CURC Matlab currently does not support parallelization across nodes, only across cores on one node.