snappyHexMesh parallel speedup?
Does anyone know of any case studies for speeding up snappy in parallel? I am getting OpenMPI errors when I try to run on more than one node and I was wondering if it's worth it to spend significant time troubleshooting.
UPDATE: Looks like I can get it to run on multiple nodes, so long as I don't use every core on each. Speedup seems pretty good so far. Went from ~5 ks on 6 processes to ~1 ks on 24 processes for a mesh ending with 8M cells. |
Hi Pete,
I use these steps and of course get a faster meshing process with SHM. 1. decomposePar 2. foamJob -p -s snappyHexMesh 3. reconstructParMesh -mergeTol 1e-06 - latestTime :) Good Luck |
All times are GMT -4. The time now is 10:14. |