On Jan 4, 2012, at 3:20 AM, Ferenc Tasnadi wrote: > Fine. After that I wanted to try the diffusion/mesh20x20.py with using > FIPY_SOLVERS=Trilinos, but > it works only if the mesh is not larger than 10x10. In case I chose a larger > one I get segmentation fault (signal 11). > With Pysparse on 1 CPU the script (mesh20x20.py) works just fine with large > meshes.
Have you tried with fewer than 16 processors? FiPy parallelizes the Grid meshes by cutting them into slabs and by the time it slices 20 cells into 16 groups, some nodes don't end up having any cells on them at all. I suspect that this is what's causing the problem. FiPy should be more robust to this situation (and I thought that we tested that it was), but when I try to run diffusion/mesh20x20.py with 16 processors, FiPy tries to launch the Mayavi viewer instead of Matplotlib, which tells me that it doesn't understand that it has 2D meshes. Reduce the number of processors (8 worked for me) and see if that helps. In the next release of FiPy, you will be able to use Gmsh to obtain more efficient partitioning and should be able to assign many more nodes to a 20x20 mesh (although there's probably not much benefit in doing so; we find that the Grid slabs actually perform quite well). > Do you have any idea what went wrong during the installation and how to fix > it? I am not an expert in > installing trilinos. I use python2.7 and trilinos-10.8.3. I doubt anything is wrong with your PyTrilinos installation, but please let us know if you can't solve diffusion/mesh20x20.py on a smaller number of nodes (start with 2). _______________________________________________ fipy mailing list [email protected] http://www.ctcms.nist.gov/fipy [ NIST internal ONLY: https://email.nist.gov/mailman/listinfo/fipy ]
