Ludovic, in order to figure out which interconnect is used, you can
mpirun --mca pml_base_verbose 10 --mca mtl_base_verbose 10 --mca btl_base_verbose 10 ... the output might be a bit verbose, so here are a few tips on how to get it step by step first, mpirun --mca pml_base_verbose 10 ... in order to figure out which pml component is used. there are typically 3 components (by decreasing default priority) - pml/ucx : UCX will be used for all point to point communications - pml/cm: one mtl component will be used for all point to point communications - pml/ob1: one or more btl will be used for point to point communications. PSM and PSM2 are generally faster with mtl/psm and mtl/psm2, and they are only used by pml/cm if pml/ucx is a match, you can either blacklist it or give it a lower priority so pml/cm is picked mpirun --mca pml ^ucx ... mpirun --mca pml_ucx_priority 1 ... or force pml/cm mpirun --mca pml cm ... note there might be an option to tell UCX it should not try to do PSM/PSM2, and in that case, pml/ucx would not be selected on an Infinipath/Omnipath network. if pml/ob1 is used, a btl component is used on a pair basis. For example, btl/vader will be used for intra node communications (as long as you do no MPI_Comm_spawn(), and btl/tcp will be used for inter node communications over TCP/IP. Cheers, Gilles On Fri, Dec 20, 2019 at 7:22 PM Ludovic Courtès via users <users@lists.open-mpi.org> wrote: > > Hello, > > I’ve written about our experience packaging Open MPI for GNU Guix in a > way that works out-of-the-box on different high-speed interconnects: > > > https://hpc.guix.info/blog/2019/12/optimized-and-portable-open-mpi-packaging/ > > I’m new to the list and in fact quite new to Open MPI, so I’d love to > read the feedback and suggestions you may have! > > Thanks, > Ludo’.
