Ansys Insight: Determine the optimum resource configuration to run Lumerical FDTD simulations

LitoLito Ansys Employee Posts: 37
edited January 13 in Ansys Products

When running FDTD simulations on a local desktop, the default setting is to use the maximum number of cores in the system. Depending on your simulation, this might run slower using all cores/processors. This post shows how to determine the optimum number of processes to run FDTD simulations. Also useful when requesting for resources on the cluster to run your simulation job via the job scheduler.

How to determine the optimum resources to run FDTD simulations.

  • Open the simulation/project
  • Check your simulation memory requirements
  • Make sure that you have enough RAM/Memory installed on your machine to run the simulation.
  • If your simulation requires more than the memory installed on your machine, expect the simulation to run slow since it will use your virtual memory (hard drive) to run your simulation. Otherwise, change your simulation to lower the memory requirements or run the simulation on a different machine with sufficient memory.

Hardware/system resources used in this example:

  • 24 Logical Cores
  • 64gb RAM

Regular or small simulations using less than 2 GB of RAM:

  • Open Resources. 
  • Start with no more than 8 processes keeping threads to 1.
  • Save and run your simulation.
  • This is often the optimum configuration for most simulations.

For simulations that require more RAM:

  • Do the same process above.
  • Save and run your simulation.
  • When job details indicates the simulation is done initializing and meshing and it is in the Running state, "Quit and save".
  • A log file will be created for your simulation, [filename]_p0.log. Open the log file with a text editor. At the end of the file, it will indicate the FDTD solver speed in Mnodes/s.
  • Repeat the process while increasing the number of process until you are using the maximum available cores on your machine. e.g. 12, 18, 24, etc... (even numbers).
  • Based on the results on the simulation log file, use the process that provided the best performance (highest Mnodes/s) to run your simulation.

    NOTE: The number of processes will depend on your simulation and hardware and could vary for each simulation and machine that will run the simulation.

This discussion has been closed.