Re: [Boost-users] mpi isend to group

The problem is that you never seem to call MPI_Wait() for the requests generated by MPI_Isend (or Boost MPI's isend(), which is in fact "just" a convenient wrapper). Let me quote the the man page:

"Nonblocking calls allocate a communication request object"

These requests are only freed after calling MPI_Wait for them. Hence the memory leak.

Thanks for this information. I'm using this isend structure on some more places, so I need a release of the communication object. Phil