GridMatlab Environment Variables

On the SuperCloud system, pMatlab jobs are launched through gridMatlab, which integrates the scheduler and pMatlab in order to provide/manage the resources requested by pMatlab jobs dynamically. In order to control the job launch and execution behavior through gridMatlab, a number of gridMatlab environment variables are provisioned as shown below.

Those environment variables designated as "internal use only" are used only by gridMatlab. However, the other environment variables are meant to be modified by the user in order to change the default behavior of gridMatlab.

Use the following command to set an environment variable, replacing <var-name> and <value> with the desired values. <value> is typically yes, no, on or off (case insensitive).

setenv('<var-name>','<value>')

`GRIDMATLAB_CPU_TYPE`

Select compute nodes with a specific CPU type. Valid values are xeon-g6 and xeon-p8.

`GRIDMATLAB_DEBUG`

Enable display of scheduler job submission command by setting setenv('GRIDMATLAB_DEBUG','yes').

`GRIDMATLAB_DESKTOP_FILESYS_STUB`

For internal use only.

`GRIDMATLAB_GPU_REQ`

Select the type and number of GPUs. Valid values are volta:1 or volta:2.

`GRIDMATLAB_GRID_SSHKEY`

For internal use only.

`GRIDMATLAB_KEEP_TEMP_FILES`

By default, all temporary files will be deleted at the end of the run, but if this variable is set to YES, those files will not be deleted after the run.

`GRIDMATLAB_LITE`

Internal use only. Dynamically select a light MATLAB® version only if the light version is used to launch a job from a login node. The light version does not install any MATLAB® toolboxes, allowing for a much faster launch of MATLAB®. This variable is available starting with the MATLAB® 2017B version.

`GRIDMATLAB_LOCAL_FS`

By setting setenv('GRIDMATLAB_LOCAL_FS','no'), use of the local filesystem will be disabled for pMatlab operations such as aggregation (agg) and message communication (SendMsg/RedvMsg/BcastMsg) functions. Instead, the central Lustre parallel filesystem will be used, which could decrease your performance due to in increased filesystem latency. It may be helpful to set this variable to no while you are debugging your pMatlab code.

`GRIDMATLAB_MANYCORE`

By default, this variable is set to YES, enabling triples mode pMatlab job launches by default. To request non-triples mode pMatlab job launches, set this variable to NO.

`GRIDMATLAB_MEMORY_REQ`

This variable specifies the amount of RAM (in megabytes) to reserve for each gridMatlab process.

`GRIDMATLAB_MT_SLOTS`

By default each compute task consumes one slot (equivalent to one core in the SuperCloud environment). This variable allows you to specify the number of slots to be used by each compute task.

`GRIDMATLAB_SRUN`

By default, pMatlab jobs require that the requested resources be available when you launch your job. If the resources are not available the job will not launch. If instead you would like your job to wait in the queue until the resources are available, you can set the GRIDMATLAB_SRUN environment variable to YES prior to launch.

`GRIDMATLAB_USE_NOJVM`

This variable controls the -nojvm option in the MATLAB® command line option. By setting setenv('GRIDMATLAB_USE_NOJVM','yes') before executing the pRUN() command, it will add the -nojvm option to the MATLAB® command line. This will help speed up MATLAB® launch time.

`GRIDMATLAB_WAIT_TIME`

This variable is used to specify the wait time control in the execwait_LaunchFunctionOnGrid() function.