Newsgroups: comp.parallel.pvm
Path: ukc!uknet!pipex!sunic!EU.net!Germany.EU.net!xmp!dagoba.priconet.de!strauss
From: strauss@dagoba.priconet.de (Frank Strauss)
Subject: Re: futur of PVM discussion...
Message-ID: <1994May27.090248.3475@dagoba.priconet.de>
Organization: Private Site at Priconet, Braunschweig, Germany
Date: Fri, 27 May 1994 09:02:48 GMT
References: <2s1odf$btv@upsn2.u-psud.fr>
Lines: 38

harry@mesiom.obspm.fr (Olivier F. HARRY) writes:

>Hi PVMers !

>For this, we made some comparisons with other distributed computing
>environments like Linda, Express, P4, Isis... We looked at
>their functionnalities, their benchmarks and the applications which
>were created on these.

Are the results available as a public electronic paper?

>Now, WE WANT YOUR OPINION on these questions, YOUR IDEAS on how to
>make PVM better. [...] to see if PVM is only a tool for
>students or a real robust system for the better developers...

Well, I'm a student, but let me mention this: :-)

Just the other day I hacked a few small programs to ping one process
on one host from another one on another host connected via ethernet.
The two hosts were SS10 and SS20 Sun workstations. The communication
methods I used were ICMP (the usual ping), UDP sockets, TCP sockets,
PVM 3.2.6 and Express. The result was, that PVM is by far the slowest.

Of course, it's a bit unfair to compare it with pure socket applications,
but I think there could be possibilities to speed it up, although the
buffer sizes are already enlarged, aren't they? I think, the problem
is that all communication gets put through the daemons, which of course
makes the PVM more robust and easy to administrate. I wonder, if it
would be a good idea to change this. Usually I would say no, but thinking
of MPP implementations of PVM, where it's *much more* needed, it should
be possible, somehow.

Please don't get my words that critical they sound. I think PVM is
great. I use it for one and a half year (a bit irregular) and I like it.

-- 
Frank Strauss, TU Braunschweig                  Uni:    strauss@ibr.cs.tu-bs.de
Buergerstr. 13, 38118 BS, (0531)-503873        Home: strauss@dagoba.priconet.de

