Asri-unix.1198
net.works
utzoo!decvax!cca!sri-unix!mo@LBL-UNIX
Fri Apr 9 07:17:58 1982
comments
From: mo at LBL-UNIX (Mike O'Dell [system])
Interlan has a Multibus 10 Meg Ether for about $2500 I believe. Tell
the you know of the forthcoming price reduction.
As for communications, I want to pour some of the cold water of reality
on these wonderous discussions.
If you are a local network research person, you are well aware of the
incredible fiascos going on in both standards efforts (IEEE 802-ring
circus)(double pun??) and everyone building networks going off and
inventing a new access protocol. Even worse, most vendors are
inventing their own higher-level protocols, many of which are
proprietary. (Vendor lock-in is a powerful marketting tool.) I also
don't buy the statement made in this forum that with this
heterogeneity, the work to intercommunicate is linear. There is an N x
M getway problem, and gateways are VERY hard when you DON'T have to do
protocol translation. If you want systems which will outlive the
current market spasms over broadband vs ethers vs rings vs
strings-and-tin-cans, the networking must be done in a way which is
INDEPENDENT OF THE WIRE, using protocols in the public domain (not
dominated by a vendor) and which will be widely implemented. At the
moment, IP/TCP/UDP are the only candidates. While they aren't my
favorite protocol designs (everyone would change something on anything,
given the chance), they ARE being implemented on MANY machines and
systems, and they are promulgated by an organization which is so large,
it is difficult for it to change its mind. This means an
implementation will have a reasonable expected lifetime. And while a
poor implementation of TCP may reduce a 10 Meg Ether to a 200Kbaud
dribble, that is 200K more than you would have if the box at the other
end spoke something you don't.
There will be systems capable of acting as glue to bind this
heterogeneous world together. You will probably want some access to
Xerox NS because if Xerox ever releases Interpress (which appears to be
in question), you want to be able to send you document to the Laser
down the hall even though your machine speaks IP/TCP.
What this is really saying is that the world is VERY heterogeneous and
will continue to be into the forseeable future. I get very nervous
when people start talking about systems which don't realize this or
honor its impact. Any system which goes off in the corner and does yet
another design as if its the only network in the world, or the only
protocol in the world, or the only anything else in the world might as
well be written in assembler language.
Here at LBL, we have learned why the greatest evolutionary lever is
currently possessed by the tortise: he takes his shell with him
where-ever he goes. Unless an organizations plans to
get in bed with one vendor and buy only that vendor's equipment from
here to eternity, you CANNOT do things permanently wedded to one
particular system. (Even then, vendors do stop making hardware and do
stop supporting software.) Sometimes this means you can't exploit every
little frob, but we don't program in assembler either. And while the
tortise is a bit slower and not quite as thoroughly-modern as some of
this compatriots, he has outlived them all. If this sounds overly
dogmatic, it is, but the point is to question whether the evolution of
workstations, this grand ticket to a distributed world, is going to
commit the same old mistakes, only in more virulent forms.
The introduction of any new technology much include planning for is
inevitable obsolecence.
-Mike
-----------------------------------------------------------------
gopher://quux.org/ conversion by John Goerzen <
[email protected]>
of
http://communication.ucsd.edu/A-News/
This Usenet Oldnews Archive
article may be copied and distributed freely, provided:
1. There is no money collected for the text(s) of the articles.
2. The following notice remains appended to each copy:
The Usenet Oldnews Archive: Compilation Copyright (C) 1981, 1996
Bruce Jones, Henry Spencer, David Wiseman.
