[ih] Fw: TCP adoption in 1984
Jack Haverty
jack at 3kitty.org
Sun May 3 15:08:38 PDT 2026
What I remember...:
- Ginny Strazisar developed the first gateway code, running on a
PDP-11. It was used to interconnect the ARPANET and SATNET, as well as
in various tests and demos of Packet Radio (PRNETs). Virginia Travers
is the same person; Ginny changed her name when she married, as was the
tradition in those days.
- The Internet was a research project, while most everyday "user"
traffic in the late 1970s was being handled just by the ARPANET.
However, some sites had begun to rely on the SATNET connection between
the US and Europe, especially Peter Kirstein's group at University
College London (UCL) and John Law's group at Radar Signals Research
Establishment (RSRE). Their work depended on Internet connectivity
well before the ARPANET was forced to use TCP after the "cutover" in
January 1983. In effect, the European users were the first users
requiring the Internet to be operational in order to get their work
done. This was often a topic of discussion at the Internet meetings,
e.g., the one in 1980 hosted by John Laws at RSRE.
- To begin a transition of the Internet "core" to an operational status,
sometime in 1981 Vint asked me if I'd be willing to take over the
"gateway work", which was being done at BBN but in a separate Division.
The Division I worked for had already run the ARPANET for a decade and
developed tools and processes for running it as an operational
"utility". ARPA had also transitioned responsibility for ARPANET to DCA
(Defense Communications Agency), which was responsible for operations
rather than research. In addition, SATNET operations had been
transitioned to the same NOC (Network Operations Center) that ran the
ARPANET on a 24x7 basis. So the Internet operations could follow a
similar path from research to operations.
- The "gateway project" transition required contractual changes, which
took a while. The BBN QTR (Quarterly Technical Report) of late 1981
documents the event and motivation. It's available at DTIC - see
https://apps.dtic.mil/sti/html/tr/ADA108783/ where the relevant section
starts on page 47 of the PDF. Many other similar QTRs document the
early history of the Internet, from BBN and other contractors.
- at BBN, in preparation for taking responsibility for making the
gateways operational, we had assembled a team of Bob Hinden, Mike
Brescia, and Alan Sheltzer. Bob Bressler had been responsible as the
PI (Principal Investigator) on the contracts for ARPA work, and he made
me the PI instead. That had the immediate effect that I now had to
collect all the information and create subsequent QTRs, and I learned
the hard way that a late QTR meant that the government didn't send a
check to BBN to pay the bill.
- Making gateways operational required a bunch of steps. The original
code was written in BCPL, which was fine for research but somewhat slow
and inefficient running in the limits of the PDP-11s we had. So the
code was rewritten in PDP-11 assembly language.
- There were also a lot of tools that had been proven useful in
operating the ARPANET, which we had to translate into the Internet
architecture. Mechanisms such as ARPANET "Traps" became aspects of SNMP
for example. A new printer appeared in the NOC, capturing reports of
activity in the "core gateways", similar to the reports that had been
coming from IMPs throughout the ARPANET.
- Bob Hinden had recently completed implementing TCP for the "TIP" to
become a "TAC" -- the terminal concentrator used extensively in
ARPANET. I had implemented the first TCP for Unix, also running on an
11/40 where the limitations of the hardware were a challenge. However,
that PDP-11/40 used a DEC IMP11A as its 1822 interface rather than
Roland Bryan's ACC hardware, Mike Brescia and Alan Sheltzer were
experienced engineers to complete the project team.
- There was at least a hint of the effects of Licklider's vision that I
posted about recently. Bressler, Brescia, and I had all worked together
in Lick's group at MIT before migrating to BBN. Other memorable network
names were also in Lick's group in the 1970s, e.g., Abhay Bhushan (see
early FTP RFCs) and Bob Metcalfe (yes, the one who invented Ethernet).
Plus lots of students passed through doing various projects, spending
enough time to possibly become infected by Lick.
- But I'm still not sure who first plugged an Ethernet into a router
connected to some long-haul network, and became part of the neonatal
Internet. By 1983, every host that had been on the ARPANET had
implemented TCP, and workstations were becoming popular, so it likely
happened before January 1983.
/Jack Haverty
On 5/3/26 10:29, Bill Nowicki via Internet-history wrote:
> Maybe the original question was about how many were 1822 to IMPs vs. Ethernet in 1984? At Stanford, we had something like two using 1822 in the CS department (KL-10 known as SU-AI and DecSystem-20 SCORE) plus another couple in the medical center, where Bill Yeager was doing his own multi-protocol router. We ran TCP over Ethernet on probably a couple dozen Sun workstations, using the V-System, and BSD on several more Vax machines. MIT and CMU had similar environments, with only a couple machines using 1822 and the majority by then using TCP. The legacy networks were PUP at Xerox PARC and Stanford, ChaosNet at MIT, and at least at Stanford they faded away with new growth in TCP. The 3 megabit Ethernet (actually something like 2.97 to align with USA cable TV standards) was only used for Altos, the laser printer, and a batch of Unibus boards made by Xerox for PDP-11s and Vaxes.
> PUP (and I think ChaosNet?) did not require ARP, but ARP was implemented and deployed by the time the RFC was published; it is very simple. The use of repeaters and flooding bridges was only short-lived since their disadvantages were so obvious.
> Now as for any non-"research", any customer of Sun Microsystems (and the company network itself, the largest customer!) had their own TCP/IP network. However, most of these (such as Sun's) did not route IP packets to the wider Internet. There were a series of what would now be called firewall hosts, starting with Tom Lyon's SunLink product. So we had exactly one that could route to the outside Internet, and a thousand or so that routed only internally. An interesting historical question to me was: what year were the majority of TCP/IP hosts in the world not directly routable on the main research Internet? My guess would be soon after, in the mid eighties at least.
> Bill
> On Sunday, May 3, 2026 at 09:59:14 AM PDT, vinton cerf via Internet-history <internet-history at elists.isoc.org> wrote:
>
> from gemini:
>
> The first internet gateway (router), developed by Ginny Strazisar at BBN in
> 1976/1977, was a *PDP-11/40* that used the *BBN 1822* interface on both
> sides.
> The Packet Radio Side
>
> Just like the ARPANET connection, the Packet Radio Network (PRNET) side
> used a *bit-serial (Distant Host) BBN 1822 interface*.
>
> The Packet Radio Units (PRUs), manufactured by *Collins Radio* (Rockwell),
> were specifically designed to emulate an ARPANET IMP's host interface. This
> allowed existing hardware and drivers to be repurposed with minimal
> physical modification.
> Key Technical Details
>
> -
>
> *Physical Hardware:* In the PDP-11/40 gateway, the connection was
> typically handled by *ACC (Advanced Computer Communications)* interface
> cards, specifically the *ACC LH-DH/11* (Local Host-Distant Host) UNIBUS
> controller.
> -
>
> *The Interface Logic:* Even though the physical layer was the same 1822
> bit-serial interface used for the ARPANET, the *logical protocol* was
> different. Instead of the standard Host-to-IMP protocol (used to talk to
> ARPANET IMPs), the gateway used the *Channel Access Protocol (CAP)* to
> communicate with the PRU.
> -
>
> *Speed:* The 1822 interface on the PRNET side supported the radio
> network's data rates, which were either *100 kbps* or *400 kbps*
> depending on the specific PRU model and configuration.
>
> Why use 1822 for both?
>
> The use of the 1822 interface for the PRNET was a strategic choice by
> DARPA. Since the ARPANET already had a well-defined, reliable
> host-to-network interface, the developers of the PRU simply adopted the
> same physical and electrical standards. This ensured that the BBN-developed
> gateway software could use similar low-level drivers to talk to both a
> fixed landline network (ARPANET) and a mobile radio-based network (PRNET).
>
> Was there a particular reason you were looking into the hardware layer of
> the PRNET gateway—are you researching the early development of TCP/IP or
> the hardware of the PDP-11?
>
> On Sun, May 3, 2026 at 12:47 PM Barbara Denny via Internet-history <
> internet-history at elists.isoc.org> wrote:
>
>> Forgot to reply to the list.
>> ----- Forwarded Message ----- From: Barbara Denny <b_a_denny at yahoo.com>To:
>> Vint Cerf <vint at google.com>Sent: Sunday, May 3, 2026 at 09:45:36 AM
>> PDTSubject: Re: [ih] TCP adoption in 1984
>> I think for the 1976 demo the gateway was still in the packet radio
>> station and only supported ARPAnet and PRnet. I think PRUs (Packet Radio
>> Units) originally had an 1822 interface. By the time of the LPR I think the
>> interface was HDLC but that change might have happened earlier.
>> I think the 1977 demo used a separate box for the gateway and may have
>> been ported to MOS by then. The packet radio station used ELF for the
>> operating system.
>> I am waiting to hear if the MIT gateway had an ethernet interface. Quite
>> some time ago on this mailing list, Noel mentioned he found my request for
>> a gateway. I am begining to wonder if that is why i asked for one. I
>> think i had to specify exactly what interfaces i wanted.
>> barbara
>> On Sunday, May 3, 2026 at 07:40:52 AM PDT, Vint Cerf via
>> Internet-history <internet-history at elists.isoc.org> wrote:
>>
>> I don't think it needed ethernet - it was connecting PRNET to Arpanet and
>> SATNET and likely used BBN 1822 interfaces on the IMP sides. Less clear to
>> me what the interface looked like on the packet radio side - maybe a bit
>> serial link?
>>
>> v
>>
>>
>> On Sun, May 3, 2026 at 9:25 AM Carsten Bormann via Internet-history <
>> internet-history at elists.isoc.org> wrote:
>>
>>> On 2026-05-03, at 14:35, Craig Partridge via Internet-history <
>>> internet-history at elists.isoc.org> wrote:
>>>> I don't recall if it supported Ethernet or simply connected two
>>>> ARPANET-technology networks (of which there were a few at the time).
>>> https://www.internethalloffame.org/inductee/virginia-travers/
>>>
>>> …talks about “radio and satellite”.
>>> 1976/1977!
>>>
>>> Grüße, Carsten
>>>
>>> --
>>> Internet-history mailing list
>>> Internet-history at elists.isoc.org
>>> https://elists.isoc.org/mailman/listinfo/internet-history
>>> -
>>> Unsubscribe:
>>>
>> https://app.smartsheet.com/b/form/9b6ef0621638436ab0a9b23cb0668b0b?The%20list%20to%20be%20unsubscribed%20from=Internet-history
>>
>> --
>> Please send any postal/overnight deliveries to:
>> Vint Cerf
>> Google, LLC
>> 1900 Reston Metro Plaza, 16th Floor
>> Reston, VA 20190
>> +1 (571) 213 1346
>>
>>
>> until further notice
>>
>>
>> --
>> Internet-history mailing list
>> Internet-history at elists.isoc.org
>> https://elists.isoc.org/mailman/listinfo/internet-history
>> -
>> Unsubscribe:
>> https://app.smartsheet.com/b/form/9b6ef0621638436ab0a9b23cb0668b0b?The%20list%20to%20be%20unsubscribed%20from=Internet-history
>>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: OpenPGP_signature.asc
Type: application/pgp-signature
Size: 665 bytes
Desc: OpenPGP digital signature
URL: <http://elists.isoc.org/pipermail/internet-history/attachments/20260503/dcf2a4b7/attachment-0001.asc>
More information about the Internet-history
mailing list