[ih] early networking: "the solution"

[John Gilmore]

John Day via Internet-history <internet-history at elists.isoc.org> wrote:
> In the early 70s, people were trying to figure out how to interwork multiple networks of different technologies. What was the solution that was arrived at that led to the current Internet?
> I conjectured yesterday that the fundamental solution must have been in hand by the time Cerf and Kahn published their paper.
> Are you conjecturing that the solution was gateways? and hence protocol translation at the gateways?

Maybe it's too obvious in retrospect.  But the "solution" that I see was
that everyone had to move to using a protocol that was independent of
their physical medium.  That was TCP/IP.  (It could potentially have been
something else -- but the network effect of early adopters kicked in,
and has been impossible to overcome even today.)

Before TCP/IP got popular, there were protocols that ran on particular
media like PUP (on Ethernet), X.25 (on Euro telco packet networks), RJE
and BITNET (on IBM bisync leased lines), FidoNET (on dialup modems), NCP
(on Arpanet nodes), etc.  If you used that protocol, you were stuck on
that medium forever.  "Protocol translation gateways" could do some dumb
translation, but they always lost the details, because the other end
"knew" what kind of device and medium it thought it was talking to, and
the translated version of some foreign thing never actually matched the
expected model (in the nuances like timing, or support of standardized
but uncommon features).  "The solution" was for customers to learn to
NEVER use those medium-specific protocols.

So by the time I was working on stuff like this at Sun in the 1980s,
nobody serious was designing "Ethernet-only" higher level protocols.
That was too limiting, and was a classic part of the vendor lock-in
strategies that Sun was trying to break (our customers) out of by
adopting standards.  All our stuff ran above IP -- which happened to run
great over Ethernet -- but it also worked over FDDI or over SLIP or PPP
on leased or modem lines.  Or via gateways onto other media, including
radio.  It was "medium-independent".

(Sun made one mistake in that era, designing and using an Ethernet-only
bootstrap protocol called RARP, but the world largely bypassed it and
ended up using BOOTP and DHCP which were based on IP.)

(Ethernet is a great COUNTER-example; the packet format and addressing
structure is unchanged today from the DIX Ethernet standard, though the
physical medium and low level protocol have changed four to six times,
and the communication speeds have increased from 10 Mbit/sec to 800,000
Mbit/sec.  Vendors have pasted over the changes with cheap protocol
translation gateways ("Ethernet switches") that can handle the various
physical speeds and media.  But Ethernet is showing its age around the
edges.  For example, it's hard to keep an 100 Gb/s fiber Ethernet full
of 1500 byte packets, unless you are aggregating thousands of customers'
independent traffic.  Yet trying to use "jumbo frames" or even longer
packets to improve utilization or thruput when squirting terabytes of
scientific data from point A to point B over an Ethernet network is an
exercise fraught with debugging bizarre problems caused by incomplete
and incompatible implementations of this "non-standard extension" of
Ethernet in various vendors' equipment of various generations.)

The standardization of clearly defined interfaces (like IP/TCP) has had
similar cost-reducing and opportunity-enhancing network effects in other
engineering disciplines (like electrical plugs and sockets -- or even
the standardization of weights and measures).  But regional versions
sometimes develop and continue to plague global commerce.  We got lucky
that the whole globe adopted IPv4 unchanged.

	John