|January 31, 2012, 08:43||
Join Date: Dec 2010
Posts: 8Rep Power: 6
do you have any experience with the new multiLevel docomposition (core/node)? How much does it accelerate? Do you know some reference, validation, testing?
Have a nice day,
|January 31, 2012, 17:11||
Join Date: Mar 2009
Location: Lisbon, Portugal
Blog Entries: 34Rep Power: 84
The closest thing I've seen here in the forum is the n-halo decomposition: Parallelization with halo layer - so you might want to ask thibault_pringuey directly what results he got.
The other closest thing I know of is the work that is either being done or has already been done in the variant OpenFOAM 1.6-ext, where multi-threading capabilities have been (or tried to) added. I've never used it like that, so I don't have a clue if it's better or not.
The other thing you might want to look for is OpenFOAM with "OpenMP" (not OpenMPI), but I haven't seen anything about yet.