Using modules
Compute Canada 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 Compute Canada 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.
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 avail
To list the modules available on a given system, 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 complete list of all modules use the spider sub-command documented below.
Sub-command spider (Lmod only)
The spider sub-command searches the complete tree of all modules 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/1.8.4
this will display a list of the modules you must load in order to access this version.
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/4.8
could let you access the GCC compilers, version 4.8.
You can load more than one module with a single command. For example,
[name@server ~]$ module load gcc/4.8 mpi/openmpi/1.8.4
will load at the same time the GCC 4.8 compilers and the Open MPI library 1.8.4, 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
After the load sub-command, unload removes a module from your environment. For example,
[name@server ~]$ module unload gcc/4.8
would remove the GCC 4.8 compilers 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 in one go.
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.
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 the 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 only)
Lmod allows you to create a collection of modules. To do so, first load the desired modules. For example:
[name@server ~]$ module load gcc/4.8 openmpi/1.8 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 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.
Lmod vs Environment Modules
The main differences between the environment available to you on the new Compute Canada servers and the servers you have used in the past are as follows.
Module hierarchy
Most systems 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 4.8 and Open MPI 1.6, you might see modules named openmpi/1.6_gcc4.8 and fftw/3.3_gcc4.8_openmpi1.6. 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/4.8 openmpi/1.6_gcc4.8 fftw/3.3_gcc4.8_openmpi1.6
you instead use
[name@server ~]$ module load gcc/4.8 openmpi/1.6 fftw/3.3
This is made possible by using a module hierarchy. The fftw/3.3 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, Lmod 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.
Module collections
Module collections are an additional functionality provided by Lmod. See this section for more details.
Only one version loaded at a time
Lmod will refuse to load two versions of the same module at the same time. For example, you cannot have versions 4.8 and 5.4 of the GCC compilers loaded at once.
Only one module in the same family loaded at a time
It is possible for administrators to specify that two modules with different names are of the same family. Lmod will refuse to load two modules of the same family. Typical examples are compiler modules (gcc, intel), MPI modules (openmpi, mvapich2), or BLAS library modules (mkl, openblas).
Automatic replacement of modules
When Lmod 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.
Permanent modules
Lmod allows administrators to define a module as permanent or "sticky". Such a module will not be removed with the module purge command.