Path: utzoo!utgpu!news-server.csri.toronto.edu!mailrus!uunet!mcsun!ukc!dcl-cs!aber-cs!pcg
From: pcg@aber-cs.UUCP (Piercarlo Grandi)
Newsgroups: comp.arch
Subject: Re: Ethernet Factoid
Summary: The Ethernet was envisaged as a *network* not a bus.
Keywords: ethernet bandwidth perversion metcalfe
Message-ID: <1718@aber-cs.UUCP>
Date: 11 Apr 90 21:02:26 GMT
Reply-To: pcg@cs.aber.ac.uk (Piercarlo Grandi)
Organization: Dept of CS, UCW Aberystwyth
	(Disclaimer: my statements are purely personal)
Lines: 56

In article <29801@amdcad.AMD.COM> phil@pepsi.AMD.COM (Phil Ngai) writes:
  In article <76700190@p.cs.uiuc.edu> gillies@p.cs.uiuc.edu writes:

  |He complained that DEC was unreasonable in demanding 10Mb/sec
  |performance, when Xerox's 3Mb/sec performance would have been
  |perfectly adequate.

On this I thoroughly agree. The Xerox people were very clever guys, and
they were "right".

  |Who was right?  Today, PC's (Compaq systempro) are hitting the 10Mb/s
  |wall.

Only because they use the wire as if it were an IO bus or even worse a
memory bus. This is ridiculous, just as having dozens of diskless with a wire
(which has a nonlinearly worsening response to increased load!) between them
and their remote discs.

I have been told by a guy (and a very bright and competent one) that for a
large 80K fortran compile he run it on a local fast CPU and a remote large
memory CPU as disc buffer cache because so he could have the best of both
worlds; fast CPU here and large memory there. The wire was not busy
otherwise.  This is literally using an Ethernet as a 5ms. latency memory
bus. It may be advantageous at times, but in the general case...

  |How far should a standard push the state-of-the-art, given the
  |costs of doing so, the expected lifetime, and the ability to make a
  |new standard later?
  
  Xerox's system was based on CATV components. The media cost was
  extremely low. I don't know how long it took DEC's media to reach
  the point of diminishing returns in terms of economies of scale
  but certainly at my company we didn't really start to use Ethernet
  until 1985 and at least part of that was due to cost.

It is not just this. The Xerox guys designed Ethernet as a way of linking
essentially autonomous workstations together and with wider area services,
provided by servers or gateways. They never did envision the absurdity of
using it as system bus for porrly engineered loosely coupled
multiprocessors, or as IO bus or memory bus.

They saw the world as a constellation of small Ethernet segments, with
strong, *strong* locality of access, first between a workstation and its
_local_ discs, and then between a workstation and its rarely accessed peers
on the same wire, and then between a workstation and an even less frequently
accessed server or gateway on some other wire. Given such a perspective, the
performance profile of a CSMA/CD protocol are not a problem, because you
expect the wire to be rarely used, even if in those cases you want the full
bandwidth.

Metcalfe himself hints strongly at this in his interview in a recent issue of
UnixWorld.
-- 
Piercarlo "Peter" Grandi           | ARPA: pcg%cs.aber.ac.uk@nsfnet-relay.ac.uk
Dept of CS, UCW Aberystwyth        | UUCP: ...!mcvax!ukc!aber-cs!pcg
Penglais, Aberystwyth SY23 3BZ, UK | INET: pcg@cs.aber.ac.uk