Newsgroups: comp.parallel.mpi
From: gdburns@osc.edu (Greg Burns)
Subject: Re: Dynamic Spawning
Organization: Ohio Supercomputer Center
Date: 1 Dec 1994 08:36:05 -0500
Message-ID: <3bkjg5$gdj@tbag.osc.edu>

In article <3bkcpo$30u6@info2.rus.uni-stuttgart.de> zrcc0100@nfhsg2.rus.uni-stuttgart.de (Manuela Sang) writes:
>Hi,
>
>I wondered if it's planned to introduce dynamic spawning in later versions,
>but I found no hint on it.
>
>Does anybody know more about this?

The static group nature of MPI, which makes groups _much_ less stressful
to program, also makes spawning processes a non-trivial operation.  It
is unacceptable in MPI to have processes simply appear in (or leave)
your group without warning.

It is also not obvious how to specify where a new process should run.
A program that says "spawn a process on host1.edu" may not be portable to
an MPP machine.  100% source code portability is a goal of MPI.

Having said all that there are suggestions on how to orderly create
processes, for example, using inter-communicators.  I would say this
is a priority for extending MPI.  Indeed, LAM 5.3 will include a
spawn function.

The new book, "Using MPI", contains some discussion on process creation.

-=-
Greg Burns				gdburns@tbag.osc.edu
Ohio Supercomputer Center		http://www.osc.edu/lam.html

