[ih] Why did TCP win?

[John Gilmore]

Brian E Carpenter wrote:
> I've always answered that with: Because it worked, had a full suite
> of applications, and it was free on Unix, and modestly priced on
> most other operating systems.

Another factor: it was well documented, the specs were public and free,
and everyone was encouraged to read them, understand them and implement
them.  TCP/IP specs had *authors*, and many were good writers.  Other
standards were documented by *committees* and sold at rapacious markups
by monopoly vendors.

Also, DARPA did a smart thing when they funded Bill Joy at UCB to put a
working TCP/IP right into the standard BSD releases.  It led to an early
demonstration of proprietary vendors stumbling badly after they had to
compete with a well debugged free-protocol and free-software product.

I was at Sun in the early 1980s, where Ethernet and TCP/IP came built-in
to every Sun Workstation.  As Brian said, it worked.  Sun used it for
everything internally, and all of our standard software was built to
expect it.  Eventually you could get DECNET and XNS and SNA and X.25 and
all that stuff from Sun's third party vendors.  But because they were
not very popular, and were only demanded by huge organizations, they
were expensive, like maybe $500 extra *per workstation*.  (Their
development and support costs were spread over a much smaller base of
users.)  Meanwhile, anybody who just wanted their Suns to network
locally, like to share disks and printers, do email, and rlogin, used
the built-in TCP/IP.

Years later when universities (Sun's initial market) could buy worldwide
flat-rate Internet service from an NSFnet regional, TCP/IP made even
more sense.  And the edge increased by the early 1990s, when commercial
ISP's started jumping over the low government hurdles, and offering real
global Internet to anyone with money.  Uh, when I put toad.com on the
Internet in the '90s with a 56k leased line to my employer, I *saved*
money, since my variable cost uucp/usenet phone bills for relaying the
"alt" backbone went down from about $600/mo to a fixed flat rate of
about $200/mo.

Nobody's yet mentioned the typical ISP pricing model as one of TCP/IP's
reasons for success.  Telcos typically did caller-pays and variable
rates, charging all the market would bear in a monopoly dominated by
regulatory capture.  Internet service was sold as both-ends-pay, but at
flat rates that made bills predictable and affordable, without nasty
surprises.  (And if yur ISP didn't, you could find another or start your
own, that would put it out of business shortly by offering flat rates.)

ISP's in turn relied mostly on telco leased lines, which also had flat
rates, and enormous economies of scale.  E.g. a DS1 leased line at 1544
kbit/sec had 24x the capacity of a 56 kbit/sec leased line, and 1/24th
the latency, but only cost about 6x as much.  Higher speed upgrades
(e.g. DS3 at 45000 kbit/sec) had similar factors.  This meant that an
ISP who acquired a lot of customers had its per-customer backhaul costs
go DOWN.  Our little ISP, The Little Garden, eventually became the
backhaul network for dozens of smaller regional ISPs, dropping not only
our costs and our latency, but their costs and their latency, because we
passed on the savings in our prices.  This cost curve also made
upgrading bandwidth by far the cheapest way to avoid congestion.  (Due
to fiber optics, we are *still* riding that amazing curve!)

	John