M-Star comes with a queue processing interface that allows the user to build a queue of simulations and run them on GPUs. Basic resource management configuration is supported such as picking GPUs to use and number of GPUs per simulation. Loading in a parameter sweep is supported.
Configuring the queue¶
Check the GPUs you want to use for queue processing
Specify the number of GPUs per job
Select an individual case directory or a directory that contains sub-directories with the case data
The directory is scanned for cases and added to the queue
Job paths must be unique. If you attempt to add the same case more than once to the queue, the case addition will be skipped.
Select the cases in the list
Running the queue¶
Simply click Start to begin queue processing. The queue will process jobs from top of the list to the bottom. The queue gui implements simulation exclusive use of each gpu, so that only 1 job is using a given GPU at a time. When a job finishes, the next job will start up on the available resources.
If a job is queued for processing and has existing output data, you will be prompted to delete the output directories prior to starting the queue.
The queue gui will indicate a basic status about the job, whether it is queued, running, finished successfully or failed.
Double click on the job to open up the simulation log file. Note that this is not a live updating reader; it only displays the log file at a point in time.
Stopping the queue¶
Click Stop to terminate queue processing. The queue manager attempts to stop the job gracefully before force terminating the running jobs. Jobs that are not complete are marked as Queued again. Note that resuming a queue is currently not supported at this time.
Select the jobs you want to post processing and click Post-Process. This opens M-Star post with the selected case output data.