From owner-freebsd-hackers  Thu May 15 13:02:22 1997
Return-Path: <owner-hackers>
Received: (from root@localhost)
          by hub.freebsd.org (8.8.5/8.8.5) id NAA11256
          for hackers-outgoing; Thu, 15 May 1997 13:02:22 -0700 (PDT)
Received: from heathers.stdio.com (root@heathers.stdio.com [204.152.114.65])
          by hub.freebsd.org (8.8.5/8.8.5) with ESMTP id NAA11243
          for <hackers@FreeBSD.ORG>; Thu, 15 May 1997 13:02:16 -0700 (PDT)
Received: from frog (socks8.raleigh.ibm.com [204.146.167.113]) by heathers.stdio.com (8.6.12/8.6.13) with SMTP id NAA11718; Thu, 15 May 1997 13:23:23 -0400
Message-ID: <337B6B93.167E@stdio.com>
Date: Thu, 15 May 1997 16:01:23 -0400
From: Larry Lile <lile@stdio.com>
X-Mailer: Mozilla 3.01 (X11; I; AIX 1)
MIME-Version: 1.0
To: hackers@FreeBSD.ORG
CC: Nate Williams <nate@mt.sri.com>
Subject: Re: Cluster Computing in BSD
References: <199705151553.IAA22312@conceptual.com>
		<199705151725.KAA15126@phaeton.artisoft.com> <199705151815.MAA01989@rocky.mt.sri.com>
Content-Type: text/plain; charset=us-ascii
Content-Transfer-Encoding: 7bit
Sender: owner-hackers@FreeBSD.ORG
X-Loop: FreeBSD.org
Precedence: bulk

Nate Williams wrote:
> 
> > > The difference between "could" and "does" is the
> > > reason for the failure of (nearly) every business unit that sold
> > > highly parallel/cluster systems.
> >
> > Except Goodyear.  And Thinking Machines Corp.  And Cray Computing.
> > And Cray Research.  And Fujitsu.  And...
> 
> Hmm, how many of these are still in business selling highly parallel
> systems?  Sounds like failure to me...
> 
> Nate

How about IBM SP's and SP2's (read DeepBlue/DeepBlueII).  These 
machines are based on both SMP and "share nothing" aproaches.  Each
node is connected to a internal high speed network and the nodes
The "thin nodes" are just a single processor 120MHz machines, and the
"high nodes" are (up to) 8 processor SMP machines (135 MHz 604
PowerPC's).  But they do have high speed busses, up to 160 MB/s.

Their approach was to make very good SMP and single processor systems
and then link them using a "share nothing" paralell management system.
This obviously works well, just ask Kasparov.  

I admit I am biased, because I am an admin on an SP, but a lot of the
support used for these could easily be reproduced on FreeBSD machines.

For more information on the SP's:
   http://www.rs6000.ibm.com/hardware/largescale/#topic5

Larry Lile
lile@stdio.com