Newsgroups: comp.parallel.mpi
From: Nick Nevin <nevin@osc.edu>
Subject: Re: Questions about interfacing PVM and MPI
Organization: Ohio Supercomputer Center
Date: 09 Aug 1996 08:57:34 -0400
Message-ID: <vn2bugk4sw1.fsf@alex.osc.edu>


     > I would like to run both pvm and mpi programs together.  The
     > architecture of the application is the following: A Pvm program,
     > called master_pvm, spwans P tasks, named salve_mpi. Each of these
     > tasks, is both a pvm task and a mpi task.  There are two
     > problems: 1/ The call to the MPI_Init routine produces an error,
     > it seems that you cannot spawn a mpi process without the mpirun
     > command ?  Is it true ?  How can I avoid this problem ?

The LAM 6.0 implementation of MPI provides as an extension a spawn call
which allows an __MPI__ process to spawn other MPI processes. If you
used this you would have to make your master_pvm program an MPI program
and the slaves would start as MPI programs and then turn themselves into
PVM programs as well. I am not sure if the latter is possible.

     > 2/ If I succeed in solving the previous problem, how can I tell
     > to all my mpi processes that they belong to a sme group of
     > communication ?  May I use inter-communicators ?

Yes the parents and children can communicate via intercommunicators. The
LAM 6.0 spawn call returns to the parents an intercommunicator to the
children. The children can get an intercommunicator back to the
parents via a function call.

---nick.

-=-
Nick Nevin				nevin@osc.edu
Ohio Supercomputer Center		http://www.osc.edu/lam.html


