Re: Why do the mpi packages depend on rsh, etc?
Jose Marin <firstname.lastname@example.org> writes:
> On Fri, 3 Mar 2000, Mark Brown wrote:
> > MPICH is hard-coded to use /usr/bin/rsh, although it is not particularly
> > concerned about what's there and could probably be modified relatively
> > easily. It's certainly trivial to do at compile time. As far as I
> > remember, the upstream authors take the view that the performance of ssh
> > is such that you probably don't want to use it in HPC applications.
> But isn't it the case that the ssh connection is only used to execute
> shell commands, and that once they're started the mpi binaries do all
> their communications independently of the rsh/ssh connection? (i.e. no
> encryption on the SEND(), RECV(), etc.)
> In other words, one may see a slower start-up of the application, but once
> it's running you shouldn't see any difference (unless you do a lot of i/o
> to stdout, but why would you?)
I think you're right here, judging from the netstat port numbers
pertaining to my running Beowulf process. But, as you note, there is
a difference at startup, possibly also with standard output.
> Jose L Marin email@example.com
> Dept of Mathematics firstname.lastname@example.org
> Heriot-Watt University
> Edinburgh EH14 4AS, U.K.
> Phone: +44 131 451 3893
> Fax: +44 131 451 3249
> To UNSUBSCRIBE, email to email@example.com
> with a subject of "unsubscribe". Trouble? Contact firstname.lastname@example.org
Camm Maguire email@example.com
"The earth is but one country, and mankind its citizens." -- Baha'u'llah