MATLAB® is a licensed high level language and modeling toolkit. The MATLAB Compiler™ lets you share MATLAB programs as standalone applications. MATLAB Compiler is invoked with
mcc. The compiler supports most toolboxes and user-developed
interfaces. For more details, check the list of supported toolboxes
and ineligible programs.
All applications created with MATLAB Compiler use MATLAB Compiler Runtime™ (MCR), which enables royalty-free deployment and use. We assume you have access to a machine that has MATLAB compiler because the compiler is not available on OSG Connect. MATLAB Runtime is available on OSG Connect.
Although the compiled binaries are portable, they need to have a compatible execution environment in terms of the operating system and matlab version. We recommend the compilation of the program with matlab versions 2014a, 2014b, 2015a, or 2015b and on machines with Scientific Linux version 5, 6 or 7 so that the compiled binaries are portable on OSG machines.
In this tutorial, we learn the basics of compiling MATLAB programs on a licensed linux machine and running the compiled binaries using MCR on the OSG machines.
Lets start with a simple MATLAB script
hello_world.m that prints
Hello World! to standard output.
function helloworld fprintf('\n=============') fprintf('\nHello, World!\n') fprintf('=============\n') end
OSG connect does not have a license to use the MATLAB compiler. On a Linux machine with a MATLAB
license, invoke the compiler
mcc. We turn off all graphical options (
-nodisplay), disable Java (
-nojvm), and instruct MATLAB to run this application as a single-threaded application (
mcc -m -R -singleCompThread -R -nodisplay -R -nojvm hello_world.m
-m means C language translation during compilation, and the flag
-R indicates runtime options. The compilation would produce the files:
`hello_world, run_hello_world.sh, mccExcludedFiles.log` and `readme.txt`
hello_world is the standalone executable. The file
run_hello_world.sh is MATLAB generated shell script.
mccExcludedFiles.log is the log file and
readme.txt contains the information about the compilation process. We just need the standalone binary file
Running standalone binary applications on OSG
To see which releases are available on OSG:
$ ssh firstname.lastname@example.org # login on OSG connect login node $ module spider matlab ------------------------------------------------------- matlab: matlab/R2018b ------------------------------------------------------- This module can be loaded directly: module load matlab/R2018b
Let us say you have created the standalone binary
hello_world. Transfer the file
hello_world to login.osgconnect.net. Alternatively, you may also use the readily available files by invoking the
$ tutorial matlab-HelloWorld # Copies input and script files to the directory tutorial-matlab-HelloWorld.
This will create a directory
tutorial-matlab-HelloWorld. Inside the directory, you will see the following files
hello_world # compiled executable binary of hello_world.m hello_world.m # matlab program hello_world.submit # condor job description file hello_world.sh # execution script
Executing the MATLAB application binary
The compilation and execution environment need to the same. The file
hello_world is a standalone binary of the matlab program
hello_world.m which was compiled using MATLAB 2014b on a Linux platform. The login node and many of the worker nodes on OSG are based on Linux platform. In addition to the platform requirement, we also need to have the same MATLAB Runtime version.
Load the MATLAB runtime for 2014b version via module command. On the terminal prompt, type
$ module load matlab/R2018b
The above command sets up the environment to run the matlab/2014b runtime applications. Now execute the binary
$ ./hello_world (would produce the following output) ============= Hello, World! =============
If you get the above output, the binary execution is successful. Next, we see how to submit the job on a remote worker machine using HTcondor.
Job execution and submission files
Let us take a look at
Universe = vanilla # One OSG Connect vanilla, the preffered job universe is "vanilla" Executable = hello_world.sh # Job execution file which is transffered to worker machine transfer_input_files = hello_world # list of file(s) need be transffered to the remote worker machine Output = Log/job.$(Process).out⋅ # standard output Error = Log/job.$(Process).err # standard error Log = Log/job.$(Process).log # log information about job execution requirements = OSGVO_OS_STRING == "RHEL 7" && Arch == "X86_64" && HAS_MODULES == True queue 10 # Submit 10 jobs
The wrapper script
#!/bin/bash set -e module load matlab/R2018b chmod +x hello_world ./hello_world
loads the correct matlab module and executes the binary.
Before we submit the job, make sure that the directory
Log exists on the current working directory. Because HTcondor looks for
Log directory to copy the standard output, error and log files as specified in the job description file.
From your work directory, type
$ mkdir -p Log
Log directory would send the jobs to held state.
We submit the job using the
condor_submit command as follows
$ condor_submit hello_world.submit //Submit the condor job description file "hello_world.submit"
Now you have submitted an ensemble of 10 MATLAB jobs. Each job prints
hello world on the standard
output. Check the status of the submitted job,
$ condor_q username # The status of the job is printed on the screen. Here, username is your login name.
hello_world.m script sends the output to standard output. In the condor job description file, we expressed that the standard output is written on the
Log/job.$(ProcessID).out. After job completion, ten output files are produced with the
hello world message under the directory
Sure, it is not very exciting to print the same message on 10 output files. In the subsequent MATLAB examples, we see how to scale up MATLAB computation on HTC environment.
This page was updated on Aug 24, 2019 at 19:00 from tutorials/tutorial-matlab-HelloWorld/README.md.