IMHO the hybris approach (MPI+threads) is interesting in case every MPI-process has lots of local data.
If you have a cluster of quad-cores, you might either have one process per node with each process using 4 threads or put one mpi-process per core. The latter is simpler because it only requires MPI-parallelism but if the code is memory-bound and every mpi-process has much of the same data, it will be better to share this common data with all processes on the same cpu and thus use threads intra-node. toon On 11/30/07, Mark Hahn <[EMAIL PROTECTED]> wrote: > > > Many problems decompose well in large chunks that are well done > > with MPI on different nodes, and tight loops that are best done > > locally with shared memory processors. > > I think the argument against this approach is more based on practice > than principles. hybrid parallelism certainly is possible, and in > the most abstract sense makes sense. > > however, it means a lot of extra work, and it's not clear how much > benefit. > if you had an existing code or library which very efficiently used threads > to handle some big chunk of data, it might be quite simple to add some MPI > to handle big chunks in aggregate. but I imagine that would most likley > happen if you're already data-parallel, which also means embarassingly > parallel. for instance, apply this convolution to this giant image - > it would work, but it's also not very interesting (ie, you ask "then what > happens to the image? and how much time did we spend getting it > distributed > and collecting the results?") > > for more realistic cases, something like an AMR code, I suspect the code > would wind up being structured as a thread dedicated to inter data > movement, > interfacing with a thread task queue to deal with the irregularity of > intra > computations. that's a reasonably complicated piece of code I guess, > and before undertaking it, you need to ask whether a simpler model of > just one-mpi-worker-per-processor would get you similar speed but with > less effort. consider, as well, that if you go to a work-queue for > handing > bundles of work to threads, you're already doing a kind of message > passing. > > if we've learned anything about programming, I think it's that simpler > mental models are to be desired. not that MPI is ideal, of course! > just that it's simpler than MPI+OpenMP. > > -mark hahn > _______________________________________________ > Beowulf mailing list, Beowulf@beowulf.org > To change your subscription (digest mode or unsubscribe) visit > http://www.beowulf.org/mailman/listinfo/beowulf >
_______________________________________________ Beowulf mailing list, Beowulf@beowulf.org To change your subscription (digest mode or unsubscribe) visit http://www.beowulf.org/mailman/listinfo/beowulf
