[1397] in Commercialization & Privatization of the Internet
Re: CSNET: A Brief History
daemon@ATHENA.MIT.EDU (Dave Crocker)
Mon Sep 23 11:14:40 1991
To: lhl@cs.wisc.edu (L.H. Landweber)
Cc: com-priv@psi.com, ietf@isi.edu
In-Reply-To: Your message of Sun, 22 Sep 91 21:28:59 -0500.
Date: Mon, 23 Sep 91 08:13:17 PDT
From: Dave Crocker <David_Crocker@Pa.dec.com>
Larry,
I had wondered what all the meetings were, that Dave Farber was going
off to, back in those days. Thanks for providing such full detail. There
was, however, one critical meeting that I did attend and it was instructive
about beliefs about networking, back in those days. Not sure where it
fits into the chronology that you gave, though.
There was a one-day meeting, held at one of the airline club meeting
rooms in Philadelphia, attended by you, Corbato of MIT, Curtis (or Adrion)
of NSF, Dave F and myself, and -- again -- others whose name excape me.
I had been told that the purpose of the meeting was to find a practical
way to make the idea of CSNet work. To be frank, the day was a set-piece.
Dave and I walked in knowing exactly what the correct answer would be,
based on the extensive experience we already had using MMDF for
relaying mail of assorted Army sites and ourselves.
The meeting began at 9am and the discussion explored the use of commercial
X.25 nets, such as Telenet. (Here comes the perceptual disconnect:) The
general belief in the room -- and elsewhere in the community -- was that
such networks really were like the Arpanet. Over the next six hours,
Dave Farber's and my contribution to the discussion primarily involved
explaining that the similarities were in the underlying technology, but
NOT the range of user functionality that was then available.
Public packet nets then -- and to a large extent today -- were used for
little more than terminal concentration. Virtually all of the sessions
were the equivalent of telnet (in 1979, X.29/X.25 wasn't operational, but
equivalent functionality was.) No general-purpose mail, file transfer,
or other inter-organization, inter-machine protocols existed on those
nets. In fact, X.25 was only marginally a standard. And to be fair,
I believe that the statistics of Arpanet usage, in those days, was very
heavily skewed towards telnet sessions.
The group became quite upset, over the course of the day, with this
bleak state of affairs and around 3pm someone finally verbally threw up
their hands and asked if there was ANYTHING that we could pursue.
I took my cue from Dave Farber and then launched into the description of
the year, or so, experience we had with MMDF and what it would take to
apply it to CSNet.
Side note about public nets: UWisc got a connection to Telenet and
I set up MMDF to start accessing them by calling up a local Telenet
terminal access line. That is, the MMDF phonenet link ran on top
of Telenet's X.29-like service, on top of X.25. Performance was ok,
but the first bill knocked everyone's socks off.
It turned out that the MMDF phonenet protocol had a link-level (we didn't
have the OSI 7-layer model back then, but the concepts still held)
protocol that broke the data into chunks of about 100 characters but
Telenet's default maximum packet size was less than that. So,
every Phonenet packet consumed TWO Telenet packets. And Telenet charged
by the packet.
Tuning the connection to make things one-for-one reduced the cost
nicely, to be less than a long-distance phone charge, but it also gave
an effective throughput of one-half the 1200 baud speed of the line,
since Telenet, of course, is store and forward and the Phonenet protocol,
then, was a classically lock-step, one data packet, one ack, exchange.
I added two-packet windowing as a direct result of this painful state of
affairs.
The lesson to this graduate student was to assume that connecting two
systems with independent history requires an additional engineering step,
beyond the simple design and implementation of a 'connector' function, and
that is to assume that there are hidden, unwanted side-effects that need
to be sought and fixed.
All in all, those definitely were interesting days. In addition to the
direct efforts of those that created CSnet, per Larry's note, I
was continually impressed with the enthusiasm and patience of the
early user sites. It usually took a day of someone's (and often only
a professor was available) time to get things running at the user site,
and then, I'm afraid, there was a constant need to work with Mike O'Brien
or me to figure out how MMDF was messing up. For all of that unfortunate
frustration, the user sites remained quite happy to have the connectivity.
Dave
