NetNews Usenet Archive 1992 #16/NN_1992_16.iso/spool/comp/dcom/cellrel/277

Text File | 1992-07-29 | 5.2 KB | 121 lines

Newsgroups: comp.dcom.cell-relay
Path: sparky!uunet!darwin.sura.net!mips!odin!sgi!rhyolite!vjs
From: [e-mail] (Vernon Schryver)
Subject: Re: Curious attitude ...
Message-ID: <[e-mail]>
Summary: fooey
Organization: Silicon Graphics, Inc.  Mountain View, CA
References: <[e-mail]> <12409@pinard> <[e-mail]> <12430@janice>
Date: Wed, 29 Jul 1992 18:38:11 GMT
Lines: 109

In article <12430@janice>, [e-mail] (Tom Gray) writes:
> ...
> Now the LAN  industry itself is rapidly accepting ATM. This
> industry expects ATM to supplant existing protocols by
> 1995. This is occuring because of the intractable management
> problems with the multitude of existing protocols. ATM
> is seen as the means for creating a workable network
> management system. I refer you to the paper
> Conquering Complexity by John Hart of 3Com which
> is in the proceedings of the 1990 IEEE Local Computer
> Network Conference which makes these points.
>
> To repeat the point, the manufacturers of LAN equipment
> regard ATM as the near term technology for their
> business. Does this make a point that ATM can be
> used profitably and efficiently for LAN problems.

Your first sentence may be close, that ATM will supplement, not
supplant, other LAN protocols by 1995, but you're off the beam for the
rest.

The reason we care (I work for a company in "the LAN industry") about
ATM is that it may be faster and cheaper than FDDI and HIPPI.  FDDI is
only 100Mb/s, and HIPPI will not have the cost advantages of high
volume production that the telephony guys might (or might not) give
ATM.

In no way will ATM ever "supplant" ethernet.  Ethernet will is the
"Model A" or "Model T" of LAN's.  It is too simple and reliable and
robust and common.  Ethernet will survive just as the Model A out
lasted the Pierce Arrow or the Cord, either of which any reasonable
person would prefer.

You restricted the issue to Local Area Networs.  I have no idea how ATM
will do in wide area or metropolitan networks or for telephony.  I
think ATM will be more popular in new networks than FDDI in 5 years
(i.e.  1997), but ATM is not a sure bet.  There are hard and unresolved
issues in putting data over ATM circuits.  The existing demo
installations are actually quite discouraging.  If you look only at how
those installations work, without being very generously hopeful about
future developments, you would write ATM off as a bad idea, and stick
to HIPPI.


> Voice has been carried in statistically multiplexed networks
> since the 1950's. These networks have well known problems
> of absolute delay, delay skew, delay jitter etc. which
> make them unsuitable for creating multiservice networks.
> Data networks such as the Internet work with routers which
> are far too costly and unreliable to provide adequate
> voice service.
>
> Now voice can be carried over the Internet if the requirements
> aren't particularly demanding in terms of performance and
> cost. My Ford Tempo can carry packages but no one would
> use it in the long haul transport business.

Gee, do you work for a phone company?
There is not enough of technical substance in those paragraphs.  You
would be more convincing if you omitted the propaganda.  There are
problems with using "statistically multiplexed networks", but isn't the
current telephone network a "statistically multiplexed network"?

> ...
> Additionally ATM is being designed to provide for
> protocols that can optimize the use of bandwidth. It
> is being designed so that adaptation layers can be
> provided to supply load smoothing (Fast Reservation,
> String, Burst Announcing etc.).

This is not the consensus of the experts on congestion problems in
networks carrying bursty traffic.  The IETF guys have been laughing at
the simplistic, peak-bandwidth allocating or hopeful algorithms (as in
"hope it works out") used by the telephony ATM guys.

Congestion avoidance and recovery are hard.  There has been a lot of
hard though and (much more important) a lot of experience in the
Internet.  No one knows how wide area ATM networks will work for
data.


> ...
> As I have pointed out before ATM is not just a link layer protocol.
> It is a family of protocols that are being designed to create
> multi-service facilities that can be deployed practically.
> As I have pointed out, it is now the protocol of choice for
> the LAN industry which expects ATM to supplant existing
> protocls by 1995.


The "LAN industry" expects no such thing.  That statement is only a
political slogan of the "ATM industry".  You'd be more convincing if
you would not simply repeat party lines.  I took the OSI-ISO pledge to
see ISO protocols replace TCP/IP in 1983.  History has proven that
silly and naive.  "ATM will supplant existing protocols by 1995" is
more obviously silly than the ISO stuff ever was.  The ATM guys don't
even have a GOSIP to waive!

It is obvious that new FDDI, token ring, and ethernet hardware will be
sold for existing networks in 1995.  It is obvious that new ethernets
will be created in 1995, if only because the cost of an ethernet today
is about $180/station.  How much will the smallest ATM switch cost in
1995?  Less than $30,000?  How much will an adapter cost in 1995?
Maybe about $2500 if you're lucky?

Get real!


Vernon Schryver,  [e-mail]