Learning Forum

[naveen.kumar]

Hi,

I and trying to simulate a rough rectangular waveguide. The structure generation and FDTD setup script is not taking more than a minute to run indiviually but while running the simulations, it is stuck on initialization status for more than half an hour. My CPU usage is less than 5% and there is ample amount of free RAM (free of any other tasks). Any particular reason for it? How can I resolve such issues?

I have a workstation with threadripper 1950 (16 cores and 32 threads). While running simulations, in the resources section; I cant change the the processes (fixed to 1) and only change the threads and capacity. Any way to resolve this.

I tried running the same simulation with my windows laptop (intel i9 13th Gen, 24 cores and 32 threads). The combination of Processes = 8 and threads = 4 is taking the least time of approx. 1hr and 30 min. If I change any of the processes and threads values, it takes longer period of time. It is not clear from the link provided on resource usage on ansys/lumerical website. Is there any way to figure out the efficient way to use the resource based on the task?

Thanks

[Lito]

Hi Naveen,

(1) If the simulation is stuck at the initialization stage, please try to remove the Lumerical Product Preference.ini files from your user's folder on your system. See KB: <<Ansys Lumerical product preference files – Ansys Optics>>

(2) On Linux, we do not bundle anymore the MPI package starting with the 2024 releases. To run on multiple MPI processes, this will require the MPI package on Linux e.g. OpenMPI or IntelMPI.  See this KB for details: <<Running simulations with MPI on Linux – Ansys Optics>>. Otherwise, without MPI, you can only run with "Local Computer" as the "Job launching preset", and can only set the number of threads up to the number of threads/logical processors on the system. The processes field will be disable since this is only for MPI. <<Resource configuration elements and controls – Ansys Optics>>. The capacity is only used running concurrent parametric sweeps. <<Concurrent Parametric Computing – Ansys Optics>> 

(3) Typically we run 1 MPI process on 1 CPU core/logical processor/thread - depending if the system has hyperthreading enabled.

• You can set the [processes x threads x capacity] up to a maximum number of cores or logical processors/threads on your machine.  
• Try different combinations as long as the product of the 3 settings does not go over 32 threads. e.g. (16x2x1), (32x1x1), (4x8x1), (2x16x1), (1x32x1)
• Using only the 24 physical cores, e.g. (24x1x1), (12x2x1), (6x4x1), (1x24x1), (2x12x1)

You will have to run with different combinations to get the optimum resource configuration for your simulation on a machine/system. This will vary depending on your simulation/model. You will have to check the 'simulation_p0.log' file for the FDTD solver speed for each resource configuration. <<Determining the optimum resources to run FDTD – Ansys Optics>>