HPC workflow in a nutshell

The "global" high-level HPC workflow is represented in the figure below. It is described more in detail in the subsequent figure and further in this manual.

Figure 5: Global HPC workflow

Main operations performed during the HPC workflow

• Creation of HPC job based on a Simcenter Amesim study.

• On-the-fly model update to newest Simcenter Amesim version, if needed.

• On-the-fly compilation of embedded user libraries, if any.

• On-the-fly compilation of the Simcenter Amesim model for the HPC platform.

• Automatic deployment of computing nodes with core fitting.

• Error tracking.

• Job monitoring.

In detail, this global workflow can be implemented in three variants, depending on the possible level of integration with Simcenter Amesim according to the end-user's specific HPC cluster setup.

Fully-integrated SSH-based workflow

The Study manager uploads model data and parameters then generates and executes a script to submit the HPC job to the scheduler from the Simcenter Amesim GUI, using an SSH connection to the HPC cluster. The HPC monitor checks the job progress and after completion it can download the saved Simcenter Amesim model with simulation results and job log files.

Figure 6: Fully-integrated SSH-based workflow

A submit script like this can be used directly by the end-user, and adapted to their own system. The script used by the HPC monitor is not shown.

Custom cluster interface

Users generate the .pack file themselves, and transfer it to the HPC cluster. They then fill in the submission template script and run it to submit the HPC job. The submission script generation can be integrated within an end-user submission system (web portal, custom submission interface…), or the command which executes the AMEHPCPreLauncher.py script can be embedded into its system.

Note

AMEHPCPreLauncher.py contains the same parameters as AMEHPCStudyStart described in the following section: AMEHPCStudystart.

Figure 7: In-house system generating and executing the submission script or calling AMEHPCPreLauncher.py

Custom job submission

Users generate the .pack file themselves, and transfer it to the HPC cluster. They then register their job in the in-house job-handling system. At job execution time, this system executes the AMEHPCStudyStart command.

Figure 8: In-house job handling system calling AMEHPCStudyStart at job execution time.

The HPC workflow uses the same communication processes and executables as the parallel computing one. It requires the use of a shared folder between the head node and the compute nodes. For more information about parallel computing in Simcenter Amesim, please consult the Parallel Processing manual.