> >> now, i.e., domain decomposition based multi-threading (not simple loop > >> splitting)? > > > > As far as I know there is no hybrid parallelization in the library at the > > moment.
But you can continue to use threads on individual nodes and MPI across nodes. step-32 can do this, for example. That said, I do not think this is a useful model because not all operations that are parallelized via MPI are also parallelized through threads. For example, deal.II can use threads (or, rather, tasks) to parallelize matrix assembly on a single MPI node, but we then hand the matrix to Trilinos or PETSc which do not use threads, only MPI. This leads to the effect that we can nicely parallelize matrix assembly if we run one MPI process on each multi-core machine, but solving the linear system is then inefficient because it uses only one core on each of these machines. Or, we run one MPI process per core to make the linear solution efficient, but then there is no point in using multiple threads per MPI process on such machines. W. ------------------------------------------------------------------------- Wolfgang Bangerth email: [email protected] www: http://www.math.tamu.edu/~bangerth/ _______________________________________________ dealii mailing list http://poisson.dealii.org/mailman/listinfo/dealii
