The or no the

Stephen Wolff, I/23/96 1:05 PM,Re: the or no the 1
From: Stephen Wolff <swolffcisco.com>
subject: Re: the or no the
To: katiehzilker.net (Katie Hafner)
Date: Tue, 23 Jan 1996 13:05:30 -0600 (EST)
Cc: swolffcisco.com (Stephen Wolff)
MIME-Version: 1.0
Katie -
Thanks for the chance to comment. Especially as it's easy to
overstate the importance of the ARPANET at the time. And I think
maybe you have. Remember, it was just ONE of a bunch of similar ideas
going on at the time - X.;25 by Merit and the telcos, DECNET - heavily
favored by the independent high-energy physics crowd,;nd M!mENet
home-grown hardware, software and protocols by Department of Energy.
They are where most of the horsepower was, and they ROUNDLY and LOUDLY
denounced TCP/IP as amateurish, insecure, and a myriad of other even
more ugly epithets.
So in fact CSNET was not formed because computer scientists wanted
access to ARPANET. The reason was in fact much more lofty and
defensible: those were early days for the discipline of computer
science; it was not even **recognized** as a discipline by the "pure"'
sciences ("Any discipline that needs 'science' in its name isn't" was
one of the gentler gibes). To be sure, quality varied widely among
academic CS departments, with (arguably) the ARPA-funded ones at the
top of the heap (many of those were funded by ONR's Marvin Denicoff,
an early CS crusader in Gummint, with ARPA money), and it was NSF's
idea (well, Larry Landweber's, Dave Farber's, and Rick Adrion's) that
if the departments were **networked** together, they could share
teaching mtls, cry on each others shoulders, pull each other up by the
bootstraps, and so on. That was the **reason** for CSNET - to improve
the breed of academic computer science departments.
It was Dave Farber, I believe, who conceived of the notion of using
the existing ARPANET to carry the long-haul traffic for CSNET, and it
was certainly he who at his political best convinced.ARPA's Bob Kahn
(another public-spirited and decidedly NOT apolitical gentleman) to
allow CSNET traffic on ARPA circuits. Difficult to overstate the
importance of this step! The FIRST case of using star multiplexing to
serve different masters. Worth a paper in itself.
Matters in the NSFNET case were very similar. The Lax Report (19787
Peter Lax NYU/Courant) commissioned by the NSF commented on the
absurdity of US scientists traveling abroad to compute on US
supercomputers, since all the supercomputers in the US were behind
barbed wire doing classified stuff, or were too expensive for mere
mortals to compute on. It went on to recommend that the NSF establish
a program under which US scientists could use US supercomputers in the
US.
Initially this would be by buying time at existing supercomputer
centers such as those of AT&T and Boeing, but that was to be only
while some NEW centers would be built (competitively) with the NSF
dollar, and which the US scientific community could use essentially
free. And oh yes by the way the Lax report said the new centers
should be networked together... NO MENTION OF THE ARPANET.
At the outset of the NSFNET program (before my time), it was
recognized that it would take time to design and build such a network,
so using the Dave Farber / CSNET example, Erich Bloch (Director of
NSF) went to Bob Kahn at ARPA and struck a deal: NSF would give ARPA
Printed for katieh@zilker.net (Katie Hafner)
------------------------------<page break>-----------------------------
Stephen Wolff, I/23/96 1:05 PM,Re: the or no the 2
(which had become DARPA by then) $4m, in return for which DARPA would
**augment** the ARPANET with 40 new sites of NSF's choosing, to allow
NSF-sponsored scientists to gain access to NSF supercomputers **over
the ARPANET**. Shared use :: second example.
Unhappily, it didn't work - since DARPA no longer actually controlled
the construction (i.e., facilities acquisition/lease) of the ARPANET;
that had been turned over to the Defense Communications Agency which
at the time was suffering from terminal bureaucratic viscosity, with
the result that the first (56 kb/s) NSFNET was built before very many
of the new ARPANET sites actually came on line, and indeed the last
of the 40 were installed and deinstalled within the same week on the
retirement of the ARPANET in June 1990.
In the end, X.25 died because it lacked end-to-end error control (it
only did link-by-link); MFENet died because (i.a) it had an address
space that limited it to 127 computers worldwide(!!); and DECNET died
because it opted for the ISO/OSI protocols which never achieved
critical mass and were in the end overwhelmed by TCP/IP coming free
with every Sun worstation and every copy of Berkeley UNIX.
Hope that helps. Cheers,
, /
Printed for katieh@zilker.net (Katie Hafner) ''
------------------------------<page break>-----------------------------