Jeff Hammond <[email protected]> writes:
You're going to have to convert to MPI+Threads eventually so you might as well start now. It is completely unreasonable to expect that MPI-only is reasonable usage on a 60-core chip with a ring interconnect.
This is something that endpoints is supposed to address. Note that OpenMP doesn't have any decent way to do reductions scoped across subsets of threads, so for many applications, it's not even a reasonable level of expressivity.
I am so convinced of the impossibility that MPI_Allreduce scales to
100 MPI procs in shared memory that I will learn Ada as punishment if I'm wrong.
Does 30 microsecond MPI_Allreduce at 512 cores qualify as "impossible to scale"? http://www.nas.nasa.gov/assets/pdf/techreports/2010/nas-10-001.pdf