rapidsai / distributed-join

Other
19 stars 12 forks source link

Use all gather instead of broadcast loop for communicating worker addresses #3

Closed nsakharnykh closed 4 years ago

nsakharnykh commented 4 years ago

Currently, nodes get addresses of remote workers through a loop of MPI_Bcast: https://github.com/rapidsai/distributed-join/blob/fc4515b6306be17e68c988a7ab6126722da342c9/src/communicator.cpp#L63-74 I think the efficiency of this code can be achieved through using MPI_Allgather once.