[limonegiallo]

Dear All,

I am trying to run a simulation with a software called CFDEMcouplig, a framework which couples LIGGGTHS and OpenFOAM. My system consists of a cylinder at the top of a truncated cone. From the bottom (inlet) to the top (outlet) of my system the air is flowing (OpenFOAM), moving several spherical particles resulting in a "particle fluidization". The particles therefore are moved by the air flow and interact between themselves and the surrounding geometry (particles interaction done by LIGGGTHS).

Now, I am trying to run this simulation on a HPC cluster with 200 to 1000 cores.
For less than 50 cores (I used 18 cores and 48 cores) everything works perfectly, but increasing the number of cores up to 200 I get an error from OpenFOAM:

[0] --> FOAM FATAL ERROR:
[0]
cfdemCloud:: local mesh has zero cells. Please check the mesh and the decomposition!

Any help or suggestion ? I need urgently to use an high number of cores for this simulation/project.


I do not know if it helps, but with the usual method to create a mesh, i.e.

cd /home/CFD
blockMesh
surfaceFeatureExtract
decomposePar
mpirun -np 4 snappyHexMesh -overwrite -parallel
reconstructParMesh -constant -fullMatch
decomposePar
mpirun -np 4 renumberMesh -overwrite -parallel

my system (cylinder at the top of a truncated cone) is split as cake slices for each processor employed. For instance, if I have/use 4 processors my system is split into 4 slices for the computation (...and I get 4 "processor" folders).

My guess, maybe totally nonsense, is that the increase of the number of processors, results in so many and "thin" slices, that for those processors there are not enough cells for the computation. Too stupid idea ?

Looking forward to hearing from you!

Best,
Limone