Newsgroups: comp.parallel.mpi
From: tomc@sloweddy.tam.uiuc.edu (tom cortese)
Subject: Non-blocking Persistent Send/Receive operations
Organization: University of Illinois at Urbana-Champaign
Date: 10 Sep 1997 21:28:22 GMT
Message-ID: <5v73dm$5la$1@vixen.cso.uiuc.edu>

	Hello again...

	I am trying to implement a non-blocking version of MPI_ALLTOALL
(so that I can mask communication latency by performing some computations)
using many non-blocking Send/Receive operations.  So far, so good.

	My question is:  Am I correct in assuming that it would be worth my
while to create persistent communication requests (since each processor will
be executing many similar Send/Receive operations) and if so, are there
non-blocking persistent communications (e.g. something analagous to
"MPI_IRECV", such as "MPI_IRECV_INIT")?

	Right now, as I see it, I can only set up BLOCKING persistent communication
operations using, for example, MPI_RECV_INIT, MPI_START_ALL, MPI_WAIT..., MPI_TEST...

	Thanks, and thanks for the encouraging responses to my earlier
notes (though I still would like to know if there are any collections of "standard"
routines written in MPI that I could browse through...)

	-Tom Cortese