SnappyHexMesh versus MeTiS
 

Hello,

I'm contemplating upgrading my OF-1.4.1 to OF-1.5 because the new meshing utility snappyHexMesh could be useful for my aerodynamics simulations. However, upon reading the UG I was left with some unanswered questions. One of them is this:

For parallel runs I use MeTiS (keyword method in the decomposeParDict) for domain decomposition because I have a need for load balancing. MeTiS can only partition 3D meshes consisting of either hexahedra or tetrahedra. Yet the new meshing utility snapphyHexMesh will/may create split-hexes. Since these are different than hexes, my guess is such a mesh would be beyond MeTiS' capabilities.

Is that true ?

Thanks, Martin

I am not sure but I think this
 

I am not sure but I think this should work with Metis. I have another question:
What about requirements of the input surface. This must be stl. However, what about free edges, T-Joints, ... in the input surface mesh? Does it have to be 100% clean and watertight or only representing my geometry?
If I take a look at the motorcycle example I see that tis stl has some free edges and kconsists of different parts. However, it does not look like an direct CAD-export at all.

Regards

Martin, Where did you see t
 

Martin,

Where did you see that Metis handles particular cell types? AFAIK it partitions graphs. For cfd/fem meshs these graphs should arise from the cell/element connectivity. I don't know enough to know how they build the graphs for networking and circuits, but I think it's safe to say that they don't have any cell types in those cases!

Anyhow Metis works fine with polyhedral meshes, both in OpenFOAM and STAR-CD.

To work in parallel the decomp
 

To work in parallel the decomposition method for the mesh generator has to be parallel-aware. The methods currently available are hierarchical and parMetis afaik.

Mark, This is specified in
 

Mark,

This is specified in the manual for MeTiS 4.0 for the mesh partitioning routines (p.34-35). Whereas the graph partitioning routines obviously don't care about element types, the mesh partitioning routines are restricted as I mentioned in my original post. It seems then that OF does not employ the mesh partitioning routines of MeTiS.

Thanks for your reply.

I confirm what Eugen says. Onl
 

I confirm what Eugen says. Only parMetis and hierarchical are available for running snappyHexMesher in parallel.
However, with metis algorithm in OpenFOAM you can partition any kind of mesh. Try it!

Bye,
Francesco