Matlab¶
Matlab example 1: Serial command line use¶
Matlab is installed on DelftBlue as a software module and, unless it is absolutely crucial to display graphics, should be run from the command line. Let's take a look at a simple example.
Which toolboxes are included?
>> ver
-----------------------------------------------------------------------------------------------------
MATLAB Version: 9.11.0.1769968 (R2021b)
MATLAB License Number: 329139
Operating System: Linux 4.18.0-372.26.1.el8_6.x86_64 #1 SMP Sat Aug 27 02:44:20 EDT 2022 x86_64
Java Version: Java 1.8.0_202-b08 with Oracle Corporation Java HotSpot(TM) 64-Bit Server VM mixed mode
-----------------------------------------------------------------------------------------------------
MATLAB Version 9.11 (R2021b)
Simulink Version 10.4 (R2021b)
5G Toolbox Version 2.3 (R2021b)
AUTOSAR Blockset Version 2.5 (R2021b)
Aerospace Blockset Version 5.1 (R2021b)
Aerospace Toolbox Version 4.1 (R2021b)
Antenna Toolbox Version 5.1 (R2021b)
Audio Toolbox Version 3.1 (R2021b)
Automated Driving Toolbox Version 3.4 (R2021b)
Bioinformatics Toolbox Version 4.15.2 (R2021b)
Communications Toolbox Version 7.6 (R2021b)
Computer Vision Toolbox Version 10.1 (R2021b)
Control System Toolbox Version 10.11 (R2021b)
Curve Fitting Toolbox Version 3.6 (R2021b)
DDS Blockset Version 1.1 (R2021b)
DSP System Toolbox Version 9.13 (R2021b)
Database Toolbox Version 10.2 (R2021b)
Datafeed Toolbox Version 6.1 (R2021b)
Deep Learning HDL Toolbox Version 1.2 (R2021b)
Deep Learning Toolbox Version 14.3 (R2021b)
Econometrics Toolbox Version 5.7 (R2021b)
Embedded Coder Version 7.7 (R2021b)
Filter Design HDL Coder Version 3.1.10 (R2021b)
Financial Instruments Toolbox Version 3.3 (R2021b)
Financial Toolbox Version 6.2 (R2021b)
Fixed-Point Designer Version 7.3 (R2021b)
Fuzzy Logic Toolbox Version 2.8.2 (R2021b)
GPU Coder Version 2.2 (R2021b)
Global Optimization Toolbox Version 4.6 (R2021b)
HDL Coder Version 3.19 (R2021b)
HDL Verifier Version 6.4 (R2021b)
Image Acquisition Toolbox Version 6.5 (R2021b)
Image Processing Toolbox Version 11.4 (R2021b)
Instrument Control Toolbox Version 4.5 (R2021b)
LTE Toolbox Version 3.6 (R2021b)
Lidar Toolbox Version 2.0 (R2021b)
MATLAB Coder Version 5.3 (R2021b)
MATLAB Compiler Version 8.3 (R2021b)
MATLAB Compiler SDK Version 6.11 (R2021b)
MATLAB Report Generator Version 5.11 (R2021b)
Mapping Toolbox Version 5.2 (R2021b)
Mixed-Signal Blockset Version 2.1 (R2021b)
Model Predictive Control Toolbox Version 7.2 (R2021b)
Motor Control Blockset Version 1.3 (R2021b)
Navigation Toolbox Version 2.1 (R2021b)
Optimization Toolbox Version 9.2 (R2021b)
Parallel Computing Toolbox Version 7.5 (R2021b)
Partial Differential Equation Toolbox Version 3.7 (R2021b)
Phased Array System Toolbox Version 4.6 (R2021b)
Powertrain Blockset Version 1.10 (R2021b)
Predictive Maintenance Toolbox Version 2.4 (R2021b)
RF Blockset Version 8.2 (R2021b)
RF PCB Toolbox Version 1.0 (R2021b)
RF Toolbox Version 4.2 (R2021b)
ROS Toolbox Version 1.4 (R2021b)
Radar Toolbox Version 1.1 (R2021b)
Reinforcement Learning Toolbox Version 2.1 (R2021b)
Risk Management Toolbox Version 1.10 (R2021b)
Robotics System Toolbox Version 3.4 (R2021b)
Robust Control Toolbox Version 6.11 (R2021b)
Satellite Communications Toolbox Version 1.1 (R2021b)
Sensor Fusion and Tracking Toolbox Version 2.2 (R2021b)
SerDes Toolbox Version 2.2 (R2021b)
Signal Integrity Toolbox Version 1.0 (R2021b)
Signal Processing Toolbox Version 8.7 (R2021b)
SimBiology Version 6.2 (R2021b)
SimEvents Version 5.11 (R2021b)
Simscape Version 5.2 (R2021b)
Simscape Driveline Version 3.4 (R2021b)
Simscape Electrical Version 7.6 (R2021b)
Simscape Fluids Version 3.3 (R2021b)
Simscape Multibody Version 7.4 (R2021b)
Simulink 3D Animation Version 9.3 (R2021b)
Simulink Check Version 5.2 (R2021b)
Simulink Code Inspector Version 4.0 (R2021b)
Simulink Coder Version 9.6 (R2021b)
Simulink Compiler Version 1.3 (R2021b)
Simulink Control Design Version 6.0 (R2021b)
Simulink Coverage Version 5.3 (R2021b)
Simulink Design Optimization Version 3.10 (R2021b)
Simulink Design Verifier Version 4.6 (R2021b)
Simulink PLC Coder Version 3.5 (R2021b)
Simulink Report Generator Version 5.11 (R2021b)
Simulink Requirements Version 1.8 (R2021b)
Simulink Test Version 3.5 (R2021b)
SoC Blockset Version 1.5 (R2021b)
Stateflow Version 10.5 (R2021b)
Statistics and Machine Learning Toolbox Version 12.2 (R2021b)
Symbolic Math Toolbox Version 9.0 (R2021b)
System Composer Version 2.1 (R2021b)
System Identification Toolbox Version 9.15 (R2021b)
Text Analytics Toolbox Version 1.8 (R2021b)
UAV Toolbox Version 1.2 (R2021b)
Vehicle Dynamics Blockset Version 1.7 (R2021b)
Vehicle Network Toolbox Version 5.1 (R2021b)
Vision HDL Toolbox Version 2.4 (R2021b)
WLAN Toolbox Version 3.3 (R2021b)
Wavelet Toolbox Version 6.0 (R2021b)
Wireless HDL Toolbox Version 2.3 (R2021b)
1. Load necessary modules:
This already lets you use Matlab directly in text mode on the login node:
[<NetID>@login03 ~]$ matlab
MATLAB is selecting SOFTWARE OPENGL rendering.
< M A T L A B (R) >
Copyright 1984-2021 The MathWorks, Inc.
R2021b (9.11.0.1769968) 64-bit (glnxa64)
September 17, 2021
To get started, type doc.
For product information, visit www.mathworks.com.
>>
However, since we are working on a supercomputer, our goal is to make Matlab read and write data from/to files, all by itself. Let's learn how to do this!
2. Create an example script:
Here, we will plot a sine function to a file:
x = -2*pi:0.01:2*pi;
y = sin(x);
h = figure('Visible','off');
plot(x,y);
print(h,'-dpng','mySineWave.png');
close(h);
3. Check if this works (interactively, on a login node):
This should generate the file called mySineWave.png:
4. Now, let's do the same properly, by submitting the job to the queue:
First, create a slurm script called matlab.sh:
#!/bin/sh
#
#SBATCH --job-name="matlab_demo"
#SBATCH --partition=compute
#SBATCH --time=00:10:00
#SBATCH --ntasks=1
#SBATCH --cpus-per-task=1
#SBATCH --mem-per-cpu=1G
#SBATCH --account=research-<faculty>-<department>
module load matlab
srun matlab -batch "run('/scratch/$USER/tests/matlab/sine.m'); exit;"
Second, submit your job:
This will submit the job to the queue, and will request 1 CPU per job. As a result, this will generate the slurm-XXX.out file and the mySineWave.png file.
Matlab example 2: Parallel execution on a single node¶
The following example (parEigenLocal.m) will use 24 (locally available) CPUs in parallel:
function [elapsedTime] = test_for()
nworker = 24
myCluster = parcluster('local');
parpool(myCluster, nworker)
N = 1000;
A = zeros(N,1);
tic;
parfor i = 1 : N
E = eig(rand(100))+i;
A(i) = E(1);
end
elapsedTime = toc;
end
In order to submit this job, we use the following submission script. Please note, that we have to specify that the job is running on a single node with --nodes=1 and we have to request one extra CPU for the matlab wrapper script, e.g. if we want 24 "workers", we have to specify 25 CPUs with --ntasks=25 and --cpus-per-task=1; this means, that the maximum number of CPUs you can request is 48 and the maximum number of workers you can request is 47.
#!/bin/bash
#SBATCH --job-name="pforTest"
#SBATCH --time=00:15:00
#SBATCH --nodes=1
#SBATCH --ntasks=25
#SBATCH --cpus-per-task=1
#SBATCH --mem-per-cpu=1GB
#SBATCH --partition=compute
#SBATCH --account=research-<faculty>-<department>
module load matlab
matlab -r parEigenLocal
Slurm output file will look as follows:
< M A T L A B (R) >
Copyright 1984-2021 The MathWorks, Inc.
R2021b (9.11.0.1769968) 64-bit (glnxa64)
September 17, 2021
To get started, type doc.
For product information, visit www.mathworks.com.
nworker =
24
Starting parallel pool (parpool) using the 'local' profile ...
Connected to the parallel pool (number of workers: 24).
ans =
ProcessPool with properties:
Connected: true
NumWorkers: 24
Cluster: local
AttachedFiles: {}
AutoAddClientPath: true
IdleTimeout: 30 minutes (30 minutes remaining)
SpmdEnabled: true
ans =
0.9932
Matlab example 3: Parallel execution on multiple nodes¶
In order to run Matlab on more than one node, we have to rely on Matlab Parallel Server. What Parallel Server does, is submitting the job to the queue on your behalf, provided that the "Slurm profile" is configured properly. Then you do not need a slurm script (it will be generated automatically), but only need to specify the correct "Slurm profile" in the Matlab script itself, e.g. here for 2 nodes, using 2 CPUs per node:
function [elapsedTime] = test_for()
nworker = 4
myCluster = parcluster('SlurmProfile1');
myCluster.ResourceTemplate = '--job-name=matlab-parallel --time=00:15:00 --account=research-<faculty>-<department> --nodes=2 --ntasks-per-node=2 --cpus-per-task=1 --mem-per-cpu=1GB';
parpool(myCluster, nworker)
N = 1000;
A = zeros(N,1);
tic;
parfor i = 1 : N
E = eig(rand(100))+i;
A(i) = E(1);
end
elapsedTime = toc;
end
Of course, before executing such a script, you have to setup the Slurm profile parameters correctly. In the example above, we pass these parameters with the myCluster.ResourceTemplate command:
myCluster.ResourceTemplate = '--job-name=matlab-parallel --time=00:15:00 --account=research-<faculty>-<department> --nodes=2 --ntasks-per-node=2 --cpus-per-task=1 --mem-per-cpu=1GB';
We execute the above script without GUI:
This should invoke Matlab and spawn a new job to the queue:
The result then is:
nworker =
4
Starting parallel pool (parpool) using the 'SlurmProfile1' profile ...
Connected to the parallel pool (number of workers: 4).
ans =
ClusterPool with properties:
Connected: true
NumWorkers: 4
Cluster: SlurmProfile1
AttachedFiles: {}
AutoAddClientPath: true
IdleTimeout: 30 minutes (30 minutes remaining)
SpmdEnabled: true
EnvironmentVariables: {}
ans =
1.4681
What to do if I don't have SlurmProfile1?
-
Start Matlab's GUI on a visualization node (see below).
-
Create and configure a new Slurm profile:
-
Go to Parallel -> Parallel Preferences:
-
Create a Slurm profile:
-
Configure it with appropriate parameters (the important thing there is "Additional Slurm properties", where you should enter all your usual #SBATCH flags):
You can also verify your configuration:
When everything is configured correctly, you should see the following:
When you close Matlab, these configuration files should be saved in your
/home/$USER/.matlabfolder and you should be able to submit jobs via script, as I described earlier. -
Matlab example 4: Interactive use via GUI on a visual node¶
1. Open remote connection to the visual node
Detailed discussion about visualization nodes can be found here.
a. Connect to DelftBlue and start a "desktop" job:
As a result, you should get a message along these lines:
Reserving a GPU node (partition "visual")
SLURM job ID is 912918
Waiting for job to start running on a node
We got assigned node visual01, waiting for output...
Waiting until VNC server is ready for connection...
VNC server visual01:1 ready for use!
Check which "port number" the system allocated to you, here "visual01:1", i.e. port "1".
b. Open a tunnel to the port above
Open a separate terminal window and open a tunnel to the corresponding port:
c. Connect your VNC client to the local port corresponding to the open tunnel
or graphically:
You should be now prompted with your VNC password (not your NetID password!):
Finally, we have our remote desktop up and running:
2. Load necessary modules and run Matlab GUI
Inside of the remote desktop, open a terminal and type the following:
module load 2022r2
module load visual
module load virtualgl
module load matlab
vglrun matlab -nosoftwareopengl
You should see the following:
You are good to go now.