Newsgroups: comp.parallel.mpi
From: edemaine@ug.cs.dal.ca (Erik Demaine)
Subject: Re: MPICH process start up other than mpirun
Organization: Math, Stats & CS, Dalhousie University, Halifax, NS, Canada
Date: Sun, 30 Jul 1995 00:30:31 GMT
Message-ID: <DCI6qw.JAB@cs.dal.ca>

Lavrentios Servissoglou (servissoglou@zdv.uni-tuebingen.de) wrote:
: "Willet I Beavers Jr [Bill]" <bbeavers@ll.mit.edu> wrote:
: >... I would like to be able to start
: >one program on one cpu.  Then, sometime later, start another
: >process on a different cpu, and pass data between the two
: >processes.=01=0E

: As far as I know, the current MPI standard (MPI-1) doesn't support
: dynamic process configuration. I hope it will be supported by MPI-2,
: perhaps with MPI_SPAWN(...).

Sorry, but a slight correction:  MPI-1 was a point-to-point message-passing
standard to get things "motivated."  The current MPI (which is supported
on all "MPI systems") is MPI 1.1.  See the MPI FAQ.

Other than that, you are correct.  One reason for not including an
MPI_Spawn command is that not all parallel computers support it.  So a large
PVM-like overhead will be experienced (using IP communication instead of
native message-passing on parallel computers), or portability of MPICH-like
implementions (free, portable) will suffer.

Erik
--
Erik Demaine        || edemaine@ug.cs.dal.ca  || edemaine@fx2800.dal.ca
edemaine@cs.dal.ca  || 01ERIK@ac.dal.ca       || edemaine@is.dal.ca
URL: http://ug.cs.dal.ca:3400/~edemaine/edemaine.html
--> A glob is a round substance of thick liquid. A blob is an unformed object.

