Differences between revisions 40 and 42 (spanning 2 versions)
Revision 40 as of 2020-09-02 15:44:36
Size: 7791
Editor: davidsch
Comment:
Revision 42 as of 2020-09-02 15:46:12
Size: 7773
Editor: davidsch
Comment:
Deletions are marked like this. Additions are marked like this.
Line 96: Line 96:
 1. '''Compile''' your MATLAB program. The best approach to avoid excessive usage of MATLAB licenses is to compile your MATLAB program. The resulting executable does not need ANY MATLAB license! This frees you from all MATLAB license restrictions and also allows you to provide your executable to users without a MATLAB installation.
 1. '''Exit''' MATLAB when after finishing your calculations/ work. Only exiting matlab will really release all allocated licenses.
 1. '''Compile''' your MATLAB program. The best approach to avoid excessive usage of MATLAB licenses is to compile your MATLAB program. The resulting executable does not need ANY MATLAB licenses! This frees you from all MATLAB license restrictions and also allows you to provide your executable to users without a MATLAB installation.
 1. '''Exit''' MATLAB after finishing your work. Only exiting matlab will really release all allocated licenses.

Matlab

Matlab is a high-level language and interactive environment for numerical computation, visualization, and programming.

Start Matlab

  • Start it on Linux (example for GNOME desktop): Type "terminal" in the search field, then click the Terminal icon. Then, type in the opened terminal window, "matlab" (Enter).
  • Start it on Windows: Start menu - Matlab - Matlab. You may also use the search function, entering "matlab" in the search field.

First Steps

Matlab has a lot of tutorials by its own. You can find them in Matlab: Help - Product Help - User Guides

Matlab Compiler

  • mcc (both: UNIX and Windows) is the MATLAB Compiler. You can use the mcc command either from the MATLAB command prompt or the command-line
  • The advantage of a compiled matlab program is, that no matlab license is taken from the ETH-licenses. This means that you should prefer to compile matlab programs for which you intend to run many instances (e.g. hundreds, in a distributed computing setup for instance). If you do not compile the programs, one license is checked out on the ETH server for each Matlab instance you are running. As the overall amount of licenses on the ETH server is limited, checking out hundreds of licenses for a distributed computing job could lead to a license shortage, meaning that you and other users at ETH cannot start new matlab instances, until some allocated licenses are released.

How to use Matlab Compiler on Linux

This section shows an example on how to use MATLAB Compiler to code and build standalone applications.

Magic Square Example

A magic square is an array of integers arranged so that their sum is the same when added vertically, horizontally, or diagonally. Copy the following code in a text-file and save it as magicsquare.m:

function m = magicsquare(n)
%MAGICSQUARE generates a magic square matrix of the size
% specified by the input parameter n.
% Copyright 2003-2011 The MathWorks, Inc.

if ischar(n)
    n=str2num(n);
end
m = magic(n);
m

Important: When this program is executed in a Linux terminal, we have to use the function str2num (string to number). Arguments are used as strings, that's why they have to be converted into a number in Matlab.

Compile the magicsquare function (with the command: mcc) in a Linux terminal.

  • Open a terminal and change to the directory where you stored magicsquare.m, then enter:

    mcc -mv magicsquare.m
  • This will generate two files: magicsquare and run_magicsquare.sh

    • magicsquare is the main file of the application.

    • run_magicsquare.sh is the script to start the application. It sets up the MCR (MATLAB Compiler Runtime) environment for the current platform architecture ($ARCH) and executes the specified command.

To run the magicsquare program we give the shell script two arguments. The first is the path to the Matlab Compiler Runtime. The second argument in this example is an application-specific param for the program magicsquare itself. Important: The version of the compiler Runtime needs to be same that was used by the compiler mcc.

> ./run_magicsquare.sh /usr/pack/matlab-2019b-fg 5
------------------------------------------
Setting up environment variables
---
LD_LIBRARY_PATH is .:/usr/pack/matlab-2019b-fg/runtime/glnxa64:/usr/pack/matlab-2019b-fg/bin/glnxa64:/usr/pack/matlab-2019b-fg/sys/os/glnxa64:/usr/pack/matlab-2019b-fg/sys/java/jre/glnxa64/jre/lib/amd64/native_threads:/usr/pack/matlab-2019b-fg/sys/java/jre/glnxa64/jre/lib/amd64/server:/usr/pack/matlab-2019b-fg/sys/java/jre/glnxa64/jre/lib/amd64/client:/usr/pack/matlab-2019b-fg/sys/java/jre/glnxa64/jre/lib/amd64

m =

    17    24     1     8    15
    23     5     7    14    16
     4     6    13    20    22
    10    12    19    21     3
    11    18    25     2     9

More help on mcc: Enter on commandline: mcc -help

Or see the Matlab-Documentation in Matlab: Help - Product Help - Matlab Compiler

Matlab on the cluster (Condor, Slurm)

  • Every Matlab-Job on the cluster uses one license. Therefore its very important, to run only compiled Matlab applications on the cluster.
  • Don't expand the Matlab search path with a large number of directories on NFS-mounted resources such as your home directory. Don't use the matlab function genpath to expand the Matlab search path.

  • Set the environment variable MCR_CACHE_ROOT to a directory on your local computer (with MCR_ROOT_CACHE pointing to your (NFS-)home directory, the load on the home servers becomes too high:

     mkdir -p /tmp/$USER
     export MCR_CACHE_ROOT=/tmp/$USER

    If you get the error message: "Could not access MCRcomponent cache", you need to make sure the path where MCR_ROOT_CACHE points to is used solely by your MATLAB instance (and not other instances, e.g. from other users). Also make sure you have the right to access the path you've set with MCR_ROOT_CACHE.

  • Use the following options for mcc when compiling:

     mcc -m -R -singleCompThread -R -nodisplay -R -nojvm foo.m
  • -R -singleCompThread: Limits MATLAB to a single computational thread. By default, MATLAB makes use of the multithreading capabilities of the computer on which it runs.

  • -R -nodisplay: Do not display any X windows or X commands.

  • -R nojvm: Start MATLAB without the JVM software. Any tools that require Java software, such as the desktop/ GUI tools, cannot be used. Handle graphics and related functionality are not supported; MATLAB produces a warning when you use them. If you need to use Java software, you can run mcc without this option.

Limited amount of MATLAB licenses

The number of MATLAB licenses is limited. To ensure a fair and efficient usage of the (site-)licenses throughout the ETH, the following measures have been put in place:

  • Matlab instances being idle for more than four hours will return their allocated license to the license pool
  • Excessive usage of licenses will be monitored and users will be asked to return licenses if needed

How to avoid allocating too many MATLAB licenses?

  1. Compile your MATLAB program. The best approach to avoid excessive usage of MATLAB licenses is to compile your MATLAB program. The resulting executable does not need ANY MATLAB licenses! This frees you from all MATLAB license restrictions and also allows you to provide your executable to users without a MATLAB installation.

  2. Exit MATLAB after finishing your work. Only exiting matlab will really release all allocated licenses.

  3. Announce the need for many MATLAB licenses, e.g. due to a MATLAB course, by registering your need in advance. Countermeasures can be taken in this case by the IT department which ensure that the requested amount of licenses will available. Use the IDES contact form therefore: https://idts.ethz.ch/hdweb/ (Nethz-login required)

  4. Buy your own MATLAB licenses if you must have a guaranteed 24/7 license availability or if you need constantly an excessive amount of licenses.

How do I know which MATLAB licenses I use?

  • The MATLAB command license('inuse') lists the licenses allocated for the current MATLAB session.

How do I know how many MATLAB licenses are used by whom?

  • You can list all checked out MATLAB base licenses using the following command on Linux:

     lmutil lmstat -c 1965@lic-matlab.ethz.ch -f MATLAB | less
  • MATLAB base licenses and all Toolbox licenses are listed using the following command:

     lmutil lmstat -c 1965@lic-matlab.ethz.ch -a | less


CategoryBTCH

Programming/Matlab (last edited 2023-12-18 07:51:43 by stroth)