From: gpt@k2.pacific.llnl.gov (Greg P Tomaschke)
Newsgroups: comp.parallel.mpi
Subject: MPI Barriers and pending communications
Followup-To: comp.parallel.mpi
Date: 27 May 1999 21:51:37 GMT
Organization: Lawrence Livermore National Laboratory
Message-Id: <7ikep9$rh9$1@lll-winken.llnl.gov>
Xref: ukc comp.parallel.mpi:5138


Does a call to MPI_Barrier guarantee that all previously-sent messages
will have "arrived" when the barrier is exited?

Consider the following situation:

1) Lots of Sends are posted to lots of MPI tasks;  any particular task will
have an unknown number of messages pending.

2) call  MPI_Barrier

3) Call MPI_Iprobe/MPI_Recv until Iprobe shows there are no more messages
waiting.  

Does the call to MPI_Barrier guarantee that we won't have any outstanding
messages (generated during phase 1) arriving later?

Thanks,
Greg

Greg Tomaschke
Lawrence Livermore National Lab
gpt@k2.pacific.llnl.gov
(To send email, remove .pacific from my return address)