snappyHexMesh in parallel leads to segmentation fault
 

Hi, I'm sorry if my questions are trivial and if my post is missing important information (please tell me what is missing, if so), but I'm a real newbiew when it comes to OpenFOAM, linux and openmpi.

I have a 'simple' case where I want to do the meshing with snappyHexMesh. My starting point is a BBox (from blockMesh) and one object (from STL file) inside. I ran snappyHexMesh without the parallel option and snappyHexMesh terminated fine, without any errors.

Since a few days, I have access to a server with 88 cores, so I thought, I'll give it a go. Now, if I decompose my mesh using decomposePar (simple) and run snappyHexMesh, the proceedure crashes after a few iterations and I get a 'mpirun noticed that process rank 29 with PID 0 on node openfoam-test exited on signal 11 (Segmentation fault).' error. However, if I take the castellatedMesh result from my previous serial run and progress to run the snapping (run decomposePar and snappyHexMesh (snap true only) for the latestTime), everything works fine in parallel.

What does a 'segmentation fault' stand for? How can I get more information on this error? Does it occur because of a too simple decomposing (simple)? If so, is there a guide on how to split your mesh best?

Thanks for any hints and as I said - I'm sorry for any stupid questions & please be kind and help me providing relevant information...

decomposing was reason for error
 

Ok, I created *.vtk files for each processor decomposition (from the blockMesh result) and had a look at them in paraview. Some of them do look a bit weird (i.e. split in to two non-connecting blocks - why?) and after a few guesses of different (simple) decomposition arrangements, I found a working one.



Nevertheless, I'm still interested in what can cause a bad decomposition to fail and what would be a good way to decompose. How do you proceed if you want to manually decompose?