Learning Forum

[Scat]

I am trying to run a large FDTD simulation on a server. According to the memory report, 164MB is required for initialization and mesh, and 299.404GB is required for running the simulation. I am trying to run this simulation on a server that has 390GB of memory, however when I run it the memory of the server gets consumed almost immediately during the meshing phase, after which it starts using up the swap space after which I am forced to kill the process before it uses up all the swap space...

How come the memory report indicates that only 299GB of memory will be required, however when I try to run the simulation it seems to use more than 500GB during the meshing stage? And can this be fixed?

[Guilin Sun]

The memory check is an estimation, and how your computer allocates the memory for a specific simulation file will strongly depend on the hardware (chips) and software (os) as well as background running programs. I am not sure the memory of 390GB in your computer is from one core or several different cores, and how those cores communicate/coordinate to handle the meshing and simulation. As far as I know, one core has limited memory. You may check how many processes you have specified in the "Resource" tab:
If you specified fewer processes and their total memory is not enough it will free your computer. Please check and test. if problem still exists, please post some screenshots here.

[Scat]

Dear Gsun, thanks for your message.
The system I am trying to run this on has 28 cores, however I had the resources setup at 32 cores (see screenshot below). I believe this shouldn't be a problem with the memory?
I believe the problem is that during the meshing stage Lumerical uses a lot more memory when using the conformal meshing stage. I have found that also with smaller simulations (~200GB), Lumerical uses significantly more memory than reported in the memory report (>350GB). However when going into the running stage of the simulation the total memory usage is almost exactly the value reported in the memory report.
I ended up running the simulation using the staircase mesh, which worked (albeit it taking a very long time to finish).

[Guilin Sun]

Conformal meshing needs extra calculation and it was added later, so the memory estimation may not well consider it. I will let related person know this.

[Guilin Sun]

This is unfortunate. I will try to dig more about it and reply you if I have the answer. My guess is, the extra memory from conformal meshing is not counted as it was developed well after the memory estimation.