Does that mean, if there is a AllGatherV and assuming that every process belongs to default comm, there will n-1 Queue Pair between the collecting process and other processes ? n = total number of MPI processes.
The answer depends on multiple parameters, like number of processes, message size, etc. Some algorithms will require O(log(n)) connections other O(n). Also on OpenIB btl level per rank we create multiple QPs , not just one. To make things even more complicated :-) there are multiple types of QPs, like RC and XRC. -Pasha -- Joba On Thu, Jun 6, 2013 at 3:37 PM, Shamis, Pavel <sham...@ornl.gov<mailto:sham...@ornl.gov>> wrote: Default implementation of collectives is based on PML (p2p layer) that is implemented on top of BTL. Consequently it laverages RDMA capabilities to some extend. Pavel (Pasha) Shamis --- Computer Science Research Group Computer Science and Math Division Oak Ridge National Laboratory On Jun 6, 2013, at 1:59 PM, Jingcha Joba <pukkimon...@gmail.com<mailto:pukkimon...@gmail.com><mailto:pukkimon...@gmail.com<mailto:pukkimon...@gmail.com>>> wrote: Hi, I have a quick question. Is there an openib (in btl framework) equivalent in coll framework? I have an MPI application with gatherv and scatterv. I am wondering if I can leverage RDMA capabilities of the underlying Infiniband fabric. Thanks, -- Joba _______________________________________________ users mailing list us...@open-mpi.org<mailto:us...@open-mpi.org><mailto:us...@open-mpi.org<mailto:us...@open-mpi.org>> http://www.open-mpi.org/mailman/listinfo.cgi/users _______________________________________________ users mailing list us...@open-mpi.org<mailto:us...@open-mpi.org> http://www.open-mpi.org/mailman/listinfo.cgi/users _______________________________________________ users mailing list us...@open-mpi.org<mailto:us...@open-mpi.org> http://www.open-mpi.org/mailman/listinfo.cgi/users
