Large region casuing slow performance, how to speed up?

Submitted by keepfit on Fri, 08/16/2013 - 20:14

Hi Chris,

I created a large region compared to the radii of particles, and the time-stepping becomes very slow.

region reg block -0.01 8.1 0.0 1.51 -0.01 4.51 units box

The particles radii varies from 0.008 to 0.016, and total # of particles is 10k. So the region is mostly empty.

I tried using shrink-wrapped "boundary s s s" but the issue remains.

Does region mesh/tet help to speed up the simulation?
I tried this feature but failed: the vtk Mesh is closed surface mesh, when importing to Liggghts, it says: VTK mesh file is incompatible with simulation box: One or more vertices outside simulation box, but it shows the same as the original STL mesh in paraview. So what vtk format mesh does "region mesh/tet" need?

Update1: The width in Y has been changed to 0.3m, which becomes a "2D" problem. BUT, the speed of time-stepping remains more or less the same as the large 3D case, how come?



Image icon DEM 3D368.63 KB
Image icon DEM 2D251.49 KB

cstoltz | Fri, 08/16/2013 - 22:02


Are you running in parallel or serial? If parallel, what is the xyz processor split? It may well be that going to the 2D problem isn't changing the xyz split, in which case you won't see much of a difference in speed.


keepfit | Sun, 08/18/2013 - 16:48

Hi Chris,

I only run this case with ONE core, so the "processors x y z" is not set. And if run in parallel, the following commanded is issued because there are only 2 cores with my PC.
mpirun -np 2 liggghts -in filename

If run with CFDEM solver, the command will not pass "processors 1 2 1". Without it, the coupling between CFD (decomposePar n (1 2 1); ) and DEM (in parallel) works fine. The speed up of DEM is a bit faster.

So I don't understand if the region is much larger (in terms of total volume) than the particle radius (1000 x) and their sum Vol, the calculation will be slow down, even though the elements contained in DEM are small: particles < 10,000 and triangles of wall mesh 1k ~5k.

I feel like it's similar to CFD: if cell size is very small compared to the domain, the total cells will be huge and calculation needs more time. So how deos region influence the speed of calculation, could you please give me some tips or info about this?