How to run faster in the HPC?

MFiX
#1

Hi
I want to run a hybrid simulation on an HPC system. I want to know how to run so that the simulation can be done faster? OpenMP or MPI? With what combination?
I have 20 cores and 60 GB of RAM.
The geometry of the problem is two-dimensional and has the following dimensions.

112
112346×506 6.92 KB

Cell size = 1 mm
Number of cells = 90,000

Thank you for your bits of help.

#2

Have you tried nodesi=2 nodesj=10 nodesk=1
?
That should go fast

#3

Thank you for your reply.
Your opinion is to use DMP. Is there an advantage over the SMP?
especially for my case.

#4

I would try both and compare. Each cluster is different. I have noted gigantic speed up even in 2D using DMP over SMP.

#5

Thanks for your help. They were really useful.

#6

You might already read the papers on MFIX DMP, SMP

1.Development of parallel DEM for the open-source code MFIX
https://doi.org/10.1016/j.powtec.2012.09.006
2. Hybrid parallelism in MFIX CFD-DEM using OpenMP
https://doi.org/10.1016/j.powtec.2014.03.047

#7

The mentioned articles contained very useful content.
Thanks a lot.