Multithreading for meshing

[Anonymous]

Hi,

I am interested in using the parallel meshing option. Wiki help indicates the following.

=====================================

To improve meshing performance, the Autodesk Simulation CFD mesher leverages multiple computing cores. This reduces the amount of time required to generate large meshes, and better utilizes high performance computing hardware.

To enable multi-threaded meshing, enable this flag in the Flag Manager

mesh_multicore

The argument, n, is the number of cores to be used by the mesher.

======================================

In order to use this feature, do I have to have any additional license? If yes, how can i get that easily? Does Cloud solution uses this option if its enabled in my model set-up?

Thanks,

VD

[Anonymous]

you can add this flag in the flag manager. Makes sense for huge models. For small models the meshing process can be even slowed down. Just try a few model sizes for 1,2,4,... cores.

[Anonymous]

Thanks Marco for the quick reply.

But do I need additional License to use this feature (multi-threading in meshing)?

VD

[Anonymous · Accepted solution]

nope

[Anonymous]

Marco,

 Is there a general core utilization level (using task manager or resource monitor) that would be an indicator the multi-core function is actually slowing down the meshing process?  A couple of our guys are using this flag to set meshing to use all four cores on a 4-core 2.8GHz Xeon CPU PC with 12GB or RAM.  The system is using all four cores, but at only 25%.  Memory and other resources are barely touched.  I suspect these particular simulations are not "huge" enough to benefit and are running inefficiently at the volume meshing stage because of the multi-core setting.  Thoughts?

Thanks,

Erik

[Anonymous]

That I do not know. Just compare the run times.

If I remember rightly I've seen this slow down for models with less than 1 or 2 million elements, if using 4 cores. So for bigger models you should see a speed up.

[OmkarJ]

I saw this thread and thought of sharing a benchmarking I did recently. 

Here are the results: 

Clarifications:

1) The time was measured from pressing the Solve buttom until "Meshing complete" appeared in the message pane at the bottom. The process goes on till BC restart, Analysis completed successfully etc. but since the multithreading is used ONLY for meshing, I measured till meshing complete.

2) There were no extruded meshes and hence multithread meshing was not aborted.

3) Both the cases were done on the same part and same physics. To make sure the mesh elements remain close to 5.9 or 1.4 million, I used to coarsen by the factor of 1.05 and again refine by the factor of 0.95 for every measurement.

4) Solver computer was my own PC : 4 cores with 3.4 GHz, 16 GB RAM

5) The coarse and fine mesh was chosen to see if there is any marked benefit of multithreading for larger meshes. 

In a nutshell, I do not understand rationale behind multithreading since  I see no benefit at least in this benchmarking. I am inclined to believe that if I use cluster with 16 cores for meshing, I infact will end up slowing the process significantly. 

Guess the old ways are the best ways.

Cheers

OJ

[Anonymous]

This begs the question will clustering and using more computors significantly alter meshing calculation times?  Any experiances out there?

Mike

[Anonymous · Accepted solution]

here are the results I did with 2014 for a 6.5 mio elements test model: