[ih] Installed base momentum

[John Gilmore]

Jack Haverty via Internet-history <internet-history at elists.isoc.org> wrote:
> Networking and computing vendors were a little slower to realize TCP had won.

My perspective on this history comes from being a young "Tourist" on the
ARPAnet (via Telenet and MIT-AI), and then being employee #5 of Sun in
1982, working there for five or six years.

It was incredibly refreshing to be on the ARPAnet and have mailing lists
where standards were discussed openly, for free, among anyone who wanted
to participate.  And with drafts and standards easily available on FTP
sites, in plain ASCII text, for me to read, understand, and if I wanted
to, implement.  I had come from the IBM mainframe (APL and OS/MVT)
world.  There we had had two mainframes IN THE SAME ROOM and designed an
email system that would move messages back and forth between them with
timed transfers several times per day on manually moved 9-TRACK
MAGTAPES!  Because you really couldn't get an interface any better than
a serial port that would talk between two mainframes, and IBM's software
tended to suck.  And non-ARPAnet standards documents were obtuse,
obviously not designed to be easily understood.  Jon Postel as the RFC
author and editor completely shattered the mold there.

By the time I joined Sun in 1982, everything we did was based on
Ethernet and TCP/IP.  Our original prototypes had 3-meg Experimental
Ethernet multibus boards created by our founder, grad student Andy
Bechtolsheim.  These had been created at Stanford along with the first
8-MHz 68000 Sun CPU boards.  There might have been a tiny bit of XNS or
something around, inherited from Stanford, but everything we built and
used was IP.  Our software started with a UNIX V7 clone from Unisoft as
a stopgap.  After hiring Bill Shannon, Tom Lyon, and Bill Joy among the
first 10 employees, we rapidly moved to 4.1[abc]BSD and 4.2BSD systems
that included TCP/IP networking as a standard feature.  This took our
standalone UNIX systems and made them part of a tight network of
cooperating systems by 1983.  We adopted 3Com Multibus 10-megabit
Ethernet boards -- with separate transceivers -- as soon as they were
available.  We adopted (and I maintained) sendmail as our networked
email software.  Tons of universities adopted our hardware and software,
because we were cheaper than Vaxes, ran solid portable UNIX software,
and came with high-resolution graphics displays and high speed
networking.

The high single-unit prices of disk drives and tape drives meant that
you got much more storage per user dollar if you bought big drives and
split them among multiple workstations.  Bill Croft built a simple
block-level storage sharing protocol (nd, network disk) that ran over
UDP.  The tape access commands like dump were hacked to work simply over
TCP.  Bill and I wrote the bootstrap code that enabled diskless
workstations to boot across the Ethernet (again using UDP) from a local
server.  In parallel, Sun designed and eventually shipped and
standardized the Network File System for object-level sharing of files
and directories.  Again, it ran over IP and UDP with some TCP, on 10-meg
Ethernet.  (Bill and I also invented and published BOOTP at the time, to
automate getting your new machine's IP address.  Sun didn't ship it
then, because the NFS kernel people had picked Reverse ARP as their
bootstrap method.  BOOTP later became DHCP.)  Eventually, Sun migrated
the Ethernet interface right onto the motherboard, once we started
building larger circuit boards.

Sun had a healthy third-party catalog of applications that included OSI
and SNA and Netware and X.25 and other networking protocols.  Sun may
have even adopted some of those products under its own brand.  But they
all were licensed at something like $1000 per workstation -- real money
in 1983.  TCP/IP came free and well debugged in the basic OS.  And you
couldn't boot the machine over SNA, or share drives that way, so you had
to understand TCP/IP anyway and run it in parallel if you wanted any of
those features.

(A great short story of his early Sun history was posted last year by
Sun emp#8, Tom Lyon, as a twitter thread here:

  https://twitter.com/aka_pugs/status/1521489115585421314 )

As most of this list knows, Berkeley Unix had TCP/IP because DARPA had
funded UC Berkeley in 1978 to make it a standard feature in evolving
4.1BSD into 4.2BSD for the research community.  I don't know how few or
many dollars that investment cost them, but it got incredible bang for
the buck!  The result was that every UNIX system (except the lame
microprocessor ones like Xenix) came standard with TCP/IP.  Every
research lab that was buying minicomputers like Pyramids, DEC Vaxes,
Sequents, or workstations like Suns or SGIs, therefore had every reason
to make their whole advanced-computing network use TCP/IP.  Their email
would interoperate locally over the LAN and be gatewayed by a server
onto UUCP or BITNET or CSNET for global connectivity.  And whenever they
could beg, borrow, or steal a realtime IP connection to the wider
Internet, then poof, their local machines just popped onto the worldwide
network.  This multi-vendor consortium rapidly pushing Ethernet and
TCP/IP forward contrasted with every other wide-area networking
technology, which was either owned by one company (like DECNET or SNA)
or was managed by a slothful postal bureacracy like X.25.

The impact of that UC Berkeley funding was akin to (and prerequisite to)
the impact that came from the later research funding given to the
National Center for Supercomputing Applications, which allowed them to
create the Mosaic graphical web browser that jumpstarted public use of
the WWW (over TCP/IP).

Don't ignore the impact of Ethernet on the acceptance of TCP/IP.
Ethernet provided the first solid standardized multi-megabit-per-second
cable interface at reasonable prices.  It would support any networking
protocol, and was supported and evolved forward by hundreds of hardware
and software companies.  Ethernet was the "on-ramp" for TCP/IP, which
would have gone nowhere if it had been limited to serial port or modem
speeds (under 100 kbits/sec).  I worked where Ethernet was adopted
early, so it shortly became unremarkable to me.  Later, I suddenly knew
it had taken over the universe when I started seeing 8-pin modular jacks
and Ethernet cables on cash registers in small ordinary shops.
(Ethernet's evolution into back-end full duplex, fiber, higher speeds,
and long haul networks only cemented its position as the standard
network front end faster than modems.)  TCP/IP went to most places that
Ethernet went, in some ways riding on its coattails.

	John Gilmore