Utiliser des modules/en

Our servers can execute all software that runs under Linux. In the simplest case, the software you need will already be installed on one of the compute servers. It will then be accessible in the form of a "module". If this is not the case, you can either ask our staff to install it for you, or do it yourself.

Modules are configuration files that contain instructions for modifying your software environment. This modular architecture allows multiple versions of the same application to be installed without conflict. For new servers, modules are managed with the Lmod tool developed at TACC. This tool replaces Environment Modules, which is used on most legacy servers. If you are familiar with this system you should not be too disoriented since "Lmod" was designed to be very similar to "Environment Modules". Refer to the #Lmod vs Environment Modules section for the main differences between the two systems.

A "modulefile" contains the information needed to make an application or library available in the user's login session. Typically a module file contains instructions that modify or initialize environment variables such as PATH and LD_LIBRARY_PATH in order to use different installed programs. Note that the simple fact of loading a module doesn't execute the software in question. To learn the name of the program binary or the syntax for its use, you should read the documentation for this software. By using the module command, you shouldn't normally need to know the exact location or path of the software or library but you can nevertheless see such details about the module by means of the command module show <module-name>.

Important module commands

The command module has several subcommands. The normal syntax is

[name@server ~]$  module command [other options]

To see a list of available sub-commands, use

[name@server ~]$ module help

Sub-command spider

The spider sub-command searches the complete tree of all modules in the current standard software environment and displays it.

[name@server ~]$ module spider

If you specify the name of an application, for example with

[name@server ~]$ module spider openmpi

this will show you a list of all available versions of the application.

If you specify the name of the application along with a version number, for example with

[name@server ~]$ module spider openmpi/4.0.3

this will display a list of the modules you must load in order to access this version.

Sub-command avail

To list the modules you can load, use

[name@server ~]$ module avail

You can obtain a list of modules available for a particular library or tool, for example modules related to openmpi:

[name@server ~]$ module avail openmpi

Note that the module avail command may not list some modules that are incompatible with the modules you have loaded. To see the list of all modules other than those you've loaded and which are available for you, use the spider sub-command documented above.

Sub-command list

The sub-command list lists the modules that are currently loaded in your environment.

[name@server ~]$ module list

Sub-command load

The sub-command load lets you load a given module. For example,

[name@server ~]$ module load gcc/9.3

allows you to access the GCC compiler suite, version 9.3.

You can load more than one module with a single command. For example,

[name@server ~]$ module load gcc/9.3 openmpi/4.0

will load at the same time the GCC 9.3 compilers and the Open MPI library 4.0, compiled for GCC.

If you load a module that is incompatible with one you already have loaded, Lmod will tell you that it has replaced the old module with a new one. This can occur especially for compilers and MPI implementations.

Sub-command unload

In contrast with the load sub-command, unload removes a module from your environment. For example,

[name@server ~]$ module unload gcc/9.3

would remove the GCC 9.3 compiler suite from your environment.

If you have other modules loaded that depend on this compiler, Lmod will tell you that they have been disabled.

Sub-command purge

The sub-command purge allows you to remove all the modules you have loaded with a single command.

[nom@serveur ~]$ module purge

Some modules may be marked "sticky" (permanent) by system administrators. These will not be unloaded.

Sub-commands show, help and whatis

The sub-commands show, help and whatis provide additional information about a given module. The show sub-command displays the entire module, help displays a help message, and whatis shows a description of the module.

[nom@serveur ~]$ module help gcc/9.3

Sub-command apropos or keyword

The sub-commands apropos or keyword allow you to search for a keyword in all modules. If you don't know which module is appropriate for your calculation, you can search by description.

Loading modules automatically

We advise against loading modules automatically in your .bashrc. Instead we recommend that you load modules only when required, for example in your job scripts. To facilitate the repeated loading of a large number of modules we recommend you use a module collection.

Module collections

Lmod allows you to create a collection of modules. To do so, first load the desired modules. For example:

[name@server ~]$ module load gcc/9.3 openmpi/4.0.3 mkl

Then use the save sub-command to save this collection:

[name@server ~]$ module save my_modules

The my_modules argument is a name you give to the collection.

Then in a later session or in a job you can restore the collection with the command

[name@server ~]$ module restore my_modules

Hidden modules

Some modules are hidden. You may ignore them. They are typically modules that you don't have to load manually. They are loaded automatically based on other modules.

Module hierarchy

Many HPC clusters around the world use a flat module structure: All modules are at the same level. This becomes problematic when many combinations of versions of different modules are available on a system. For example, if you need to use the FFTW library and the module fftw is available in several versions, including a version compiled with GCC 9.3 and Open MPI 4.0, you might see modules named openmpi/4.0_gcc9.3 and fftw/3.8_gcc9.3_openmpi4.0. This is neither elegant nor practical. To solve this problem we use a hierarchy of modules. Rather than using the command

[name@server ~]$ module load gcc/9.3 openmpi/4.0_gcc9.3 fftw/3.8_gcc9.3_openmpi4.0

you instead use

[name@server ~]$ module load gcc/9.3 openmpi/4.0 fftw/3.8

This is made possible by using a module hierarchy. The fftw/3.8 module that is loaded will not be the same one that would be loaded if you had previously loaded the Intel compilers instead of GCC.

The inconvenience of using a module hierarchy is that, since modules can have the same name, only the modules that are compatible with the "parent" modules are displayed by the module avail command. Loading a parent module is therefore a prerequisite to loading some modules. To get complete information, the module system provides the module spider command. It scans the entire hierarchy and displays all the modules. By specifying a module and a particular version, it is then possible to see which paths in the hierarchy enable the desired module to be loaded.

Automatic replacement of modules

When the module system detects two modules of the same family, or two version of the same module, the command module load will automatically replace the original module with the one to be loaded. In the cases where the replaced module is a node in the module hierarchy, dependent modules will be reloaded if there are compatible versions, or deactivated otherwise.

Creating modules

For instructions about creating modules, please refer to the official documentation.

Using modules with ZSH or KSH

If you wish to use modules with the ZSH or KSH shell, execute the appropriate following command:

[name@server ~]$ zsh -l

[name@server ~]$ ksh -l

GNU Compiler Collection, an open source compiler collection

Message Passing Interface