From owner-freebsd-hackers  Thu May 15 09:47:55 1997
Return-Path: <owner-hackers>
Received: (from root@localhost)
          by hub.freebsd.org (8.8.5/8.8.5) id JAA28717
          for hackers-outgoing; Thu, 15 May 1997 09:47:55 -0700 (PDT)
Received: from apolo.biblos.unal.edu.co ([168.176.37.75])
          by hub.freebsd.org (8.8.5/8.8.5) with ESMTP id JAA28699
          for <hackers@FreeBSD.ORG>; Thu, 15 May 1997 09:47:34 -0700 (PDT)
Received: from unalmodem.usc.unal.edu.co (unalmodem15.usc.unal.edu.co [168.176.3.45])
	by apolo.biblos.unal.edu.co (8.8.5/8.8.5) with SMTP id LAA00244;
	Thu, 15 May 1997 11:47:13 -0500 (COT)
Message-ID: <337B58A8.2540@fps.biblos.unal.edu.co>
Date: Thu, 15 May 1997 11:40:40 -0700
From: "Pedro F. Giffuni" <pgiffuni@fps.biblos.unal.edu.co>
Organization: Universidad Nacional de Colombia
X-Mailer: Mozilla 3.0 (Win16; I)
MIME-Version: 1.0
To: "Russell L. Carter" <rcarter@consys.com>
CC: "Alex Fenyo (eowyn)" <fenyo@email.enst.fr>, hackers@FreeBSD.ORG
Subject: Re: Cluster Computing in BSD
References: <199705151553.IAA22312@conceptual.com>
Content-Type: text/plain; charset=us-ascii
Content-Transfer-Encoding: 7bit
Sender: owner-hackers@FreeBSD.ORG
X-Loop: FreeBSD.org
Precedence: bulk

Russell L. Carter wrote:

> > computers, but I can live with it :-). (With six boxes, a common
> > scientific process could take nearly 1/6 of the time on a fast network).
>                      ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
> 
> The difference between "could" and "does" is the
> reason for the failure of (nearly) every business unit that sold
> highly parallel/cluster systems.
> 
I already corrected this, it does 4.5 times better, not 6. Anyway
measuring this is extremely difficult and any vendor can change the
intrinsic variables.
It all depends on how big is your process(es) and if the other boxes are
busy on other things or not. I would expect better results with SMP (the
other processors work on the same thing while on clustering every box
can handle it's own process), but in the end the total cost of the
hardware is the issue, and if you already have other boxes with spare
time, well..
The type of process is also important. Clustering is better for
crunching-number process than in massive volume processes. Not much
sense in using clustering for a Web server if your internal network
isn't VERY fast.
	Pedro.

> Cheers,
> Russell
> 
> >
> >       Pedro.
> > >
> > > Alexandre Fenyo
> >
> >