[ih] TCP adoption in 1984

[Clem Cole]

below
On Sat, May 2, 2026 at 6:26 PM Bob Purvy via Internet-history <
internet-history at elists.isoc.org> wrote:

> Grok's estimate is that about 900 hosts could be reached via TCP, at the
> end of 1984. RFC 984 for TCP over Ethernet was in April of that year.
>
> Were these *entirely* university and research machines? Were any of them
> actually running TCP on Ethernet? Does anyone know"
>
Most certainly.  While many computer users did not become aware of IP/TCP
running Ethernet until after the August 1983 release of 4.2BSD Vax UNIX, other
physical networking standards and protocol stacks were available by the end
of 1980; Ethernet and IP/TCP had already been established.  And some of
those stacks ran on systems other than VAXs, much less UNIX.

I'll try to point out some key events that happened well before 1984 or RFC
984.

In late August of 1979, a "Standards Project Authorization Request" (PAR)
was submitted by Maris Graube of Tektronix and Bob Stuart (Nat Semi - I
think, but memory is a little hazy here) to the IEEE Computer Society proposing
a standard for "Local Network for Computer Interconnection".  By mid March
1980, the IEEE Standards Board had approved "Standards Project No. 802,
with Graube as the chair. (I'm going to skip some history/politics,
but the *first
draft of the IEEE version of Ethernet, 802.3, was published in 1983* (and
was finalized in 1985).

In the meantime, the original Xerox/DEC/Intel (DIX) "Bluebook" definition
of Ethernet running on what we now call 10base5 (*a.k.a.,* DIX Ethernet)
was published on *30 September 1980.*   This was what really got it started.
IEEE "standard" Ethernet would not become ubiquitous until much later, but a
lot of systems ran DIX-based systems long before the 1985 802.3 publication
date.

Nine months before that, RFC 760 DOD Standard Internet Protocol and RFC 761
DOD Standard Transmission Control Protocol were both published in *January
1980* (and in September 1981, updated and replaced by RFC 791 and 793,
respectively.

By early 1981, LANs running on top of DIX Ethernet and using an IP stack
were already being created.    I personally led the effort that started
running IP/TCP at Tektronix in very early 1981 on a few private LANs (and
the Chair of IEEE 802, while not part of CSR in TekLabs, Maris was
certainly aware of what we were doing, as his office was on the same floor,
in the same building).

Another point, Bob Metcalfe, Bruce Borden, Howard Charney, and Greg Shaw
presented their UNET IP/TCP stack for *UNIX V7 in November of 1980* [see
https://www.bitsavers.org/communications/3Com/3Com_UNET_Nov80.pdf].  (FWIW: In
an envelope, I believe I still have somewhere in my basement archives, is
the mailing envelope dated December 32, 1980, which shipped a 600 ft
9-track tape to me at Tektronix as the first customer ship of the same).

In Tek Labs, we used 3COM's original DIX-compliant 3C100 series Unibus
Ethernet interface and transceiver.   The 3Com hardware and their UNET
software were installed on Tek Labs 11/70 (which also had an NSC
Hyperchannel interface). This IP/TCP stack connected to a VMS system over
Ethernet and to a NOS-based CDC Cyber machine over Hyperchannel.  In the
beginning, all were in the same machine room. †

Meanwhile, back at BBN, Rob Gurwitz is working on the original BBN IP/TCP
implementation for 4.1BSD Vaxen.  The November 1981 release of
RFC801 includes a message dated 18-Nov 1981 from Rob that says: "Aside from
ARPANET 1822 and the ACC LH/DH-11 driver, experimental drivers have also been
developed for ETHERNET" (which were the 3COM 3C100 and Interlan NI1010).
FWIW: placed like CMU, Stanford, and MIT, which all had Altos, were running
3M Ethernet, so a driver for the Unibus 3MB-based board quickly appeared
(again, fuzzy memory, but I think MIT first wrote that).

Note that it would take another year for ARP to appear from MIT as RFC826,
as we used static host tables that mapped hostnames, IP addresses, and MAC
addresses.

Masscomp shipped its first smart Ethernet controller that ran an embedded
IP/TCP stack in early 1983.  As Tom Lyon points out, Sun shipped its first
Sun OS 1.0 with a full IP/TCP by November of 1983.  Silicon Graphics would
have been around the same time.

Going back to your original question, my experience running an IP/TCP stack
on DIX-based Ethernet was extremely well established by mid-1983 at many
sites.  How many of those sites were also Arpanet sites? That is a much
smaller number because participation was limited/controlled.

Clem


† Tek history:

Steve Glaser wrote the Hyperchannel support for the UNIX side.  The CDC IP
stack and the Hyperchannel support were written primarily by Stan Smith.
Steve, Stan, and I were the original authors of the IP stack for VMS.  The
CDC stack user code was primarily FORTRAN, with the OS working in
assembler.   VMS was primarily Bliss with some assembler.  I should point
out that the VMS driver for 3C100 VMS was written in assembler by Greg Shaw
at 3COM, with a bit of help/rewrite from Steve and me, which we fed back to
them.

I later gave Howard Wakler at CMU CS Dept a copy of the entire VMS stack
(where I had graduated the previous year).  A highly modified and enhanced
version of that stack can still be found:

   -  https://www.digiater.nl/openvms/freeware/v40/cmuip/ (src)
   - https://www.openvmshobby.com/vax-vms/carnegie-mellon-tcpip-vax/ (
   details).

 I do not know what happened to the CDC NOS stack.