Newsgroups: comp.parallel.mpi
From: impellus@ames.ucsd.edu (Tom Impelluso)
Subject: MPI-sockets (please advise)
Organization: UC San Diego, AMES Department
Date: 23 Aug 1996 23:55:46 GMT
Message-ID: <4vlge2$82f@soenews.ucsd.edu>

Hello

I recently asked a question about native Cray message passing
verse MPI.

I asked people to balance: speed vs portability-and-ease-of-use.

I need to ask a follow-up but with regard to another issue.

My work requires me to write a "server" on a remote Cray.
I must also write two "clients"  (I use quotes because those
terms are valid when using BSD sockets or TLI as an API to TCP/IP,
but MPI, it seems, can also do the job).

So, once again, I would like to ask people to advise me on:
speed vs portability-and-ease-of-use.

Should I use sockets to allow my client to connect to the remote?
Further, one of my clients is a PC upon which I installed a TCP/IP
stack to facilitate BSD sockets.  Will MPI run on that PC.

In this case, ease of writing is less important to me because I
have a some facility with socket coding.  In this case, I am more
intersted in speed.  Which is faster: sockets or MPI?

Naturally, I now know I can use MPI instead of the native Cray.
I know I could use MPI instead of sockets.
I can likely, one day, use MPI to tie my shoes in the morning.

But, at some point, I do gain some personal satisfaction out of
doing my own socket coding, and I have enjoyed learning it.
But, it seems that MPI can take the place of socket coding, too, huh?

I am h

