Hello Martin, thanks for the quick reply. You are right - the tutorial worked fine after I updated via svn and recompiled the deal.II libraries.
Thanks a lot for your help. Kind regards Florian -----Ursprüngliche Nachricht----- Von: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED] Im Auftrag von Martin Kronbichler Gesendet: Dienstag, 23. September 2008 10:48 An: dealii Betreff: Re: [deal.ii] Parallel Implementation (PETSc - Trilinos) Dear Florian, > 1. In the tutorial step-17, I just compiled the code and made > some tests with different numbers processors. I saw that the number of > required iterations to solve the linear system is strongly dependent > on the number of processors: > > #proc | required iterations in cycle 7 > > ---------------------------------------------- > > 1 | 123 > > 2 | 134 > > 8 | 151 > > What could be the reason for this behavior? The documentation of the used preconditioner used in step-17, http://www.dealii.org/developer/doxygen/deal.II/classPETScWrappers_1_1PreconditionBlockJacobi.html tells you that we use a BlockJacobi preconditioner among the blocks on the individual processors, with each processor doing an ILU preconditioning on its block. On one processor, we hence use an ILU preconditioner, whereas the parallel preconditioner gets more Jacobi-like on more processors. Since Jacobi is worse than ILU (at least for elliptic problems), there is a decrease in performance. You will generally see an increase in iterations when working in parallel, even with "good" preconditioners like multigrid (though, the increase will be less dramatic there, but there is some other overhead). > 2. In order to make a decision between PETSc and Trilinos I also > tried step-32 (the parallel version of the stokes solver). For one > processor it is no problem to run the code, but when I use more the > one, I get an error message when I call > stokes_constraints.distribute_local_to_global. Is there a problem with > my trilinos installation or doesn’t this tutorial work at the moment? Can you specify your problem? The program works at least on my system. Maybe you need to reconfigure and recompile deal.II (I changed something in the configure file last week, which causes a few undefined linker references in case configure is not re-run). Best regards, Martin _______________________________________________ _______________________________________________
