Optimizes IB transport layer for dc_alloc for DataSpaces. Uses MPI (if available) to exchange information about peer_list between ranks of same application. Clients do not write conf file to the shared storage i.e., PFS during initialization. Synchronization is inherently done via MPI_Bcast and MPI_Gather instead of sleep (in the existing code).
Optimizes IB transport layer for dc_alloc for DataSpaces. Uses MPI (if available) to exchange information about peer_list between ranks of same application. Clients do not write conf file to the shared storage i.e., PFS during initialization. Synchronization is inherently done via MPI_Bcast and MPI_Gather instead of sleep (in the existing code).