[ih] First ARPANET Link Put Into Service: November 21, 1969
Jack Haverty
jack at 3kitty.org
Sun Nov 23 17:13:20 PST 2025
Steve,
You might be interested in the behind-the-scenes story. When I was
working on a patent fight in 2012, Ben reminisced about IMP1:
"You spoke of how tight the IMP code was. Do remember that the entire
thing, buffers, data structures, and all, fit in 12k 16-bit words. I
shudder to think.
Just before we were to ship IMP 1, we had a problem where the machines
would crash mysteriously about once per 24 hours. It was a weird crash,
with the PC pointing at a completely random place, typically in a chain
of non-executable locations in a data structure, with no way it could
have gotten there. The location before would have caused a crash, there
were no jumps to it (with or without indexing ior indirection). I
concluded that it had to be a race condition with the heavy use of the
DMC channel. I went through the Honeywell drawings of the 516 processor
and found a place where the timing looked too tight. I figured a way to
patch it in the processor. I rewired it less than 24 hours before
scheduled ship. It fixed the problem and the machine shipped on
schedule. We had a heck of a time convincing Honeywell that they had
this fundamental design flaw in the central timing chain of their
machine, but they eventually were convinced and made the change in their
future machines."
Ben did a lot of the work involved in actually getting IMPs to work in
the field. Most of that has probably never been publicized, being not
as interesting as things like routing algorithms. But when we were
tasked by Vint to "make the Internet into a 24x7 operational service",
the ARPANET guys were just down the hall operating the ARPANET so it was
easy to do technology transfer to the Internet.
Ben was a hardware guy and a "doer"; here's another of his recollections
about the "reload from neighbor" capability:
"I wrote a lot of 316 code. As you say, it was pretty easy after
writing lots of assembler code for the PDP-1, PDP-8, PDP-10, and a bunch
of others. I wrote all the IMP diagnostics that I used in checking out
all the interfaces. (Did you know that the 516 IMPs were shipped with
an optical paper tape reader that was automatically activated by the
Watch Dog Timer? With a big old relay that would power the PTR up?)
That was a pain, particularly given that the paper tape had to be picked
up off the floor and rolled up again. There were 3 copies of the whole
program on the tape, but after 3 tries, you were done. And the tape
would get misplaced by site personnel. I figured out the trick of
setting up a machine to do a Modem n input over all memory and setting
up its neighbor to do the output from all memory as a MUCH better way to
reload the IMP. And you didn’t have to send out new paper tapes to all
the sites and get someone to mount them every time you did a patch."
So you can blame Ben for getting IMP1 onsite, working, early, and
destroying your grace period. I've learned that researchers rarely
think about operational issues, and the people who get it running have
to fill in the gaps - even if it means fixing someone else's design flaw
or mitigating the effects of onsite conditions.
Enjoy the history!
/Jack
On 11/23/25 15:30, Steve Crocker wrote:
> Jack,
>
> BBN was having hardware trouble as the time neared for the delivery of
> the first IMP. I was relieved because we were running late with our
> software to talk to the IMP. I figured the IMP would arrive and then
> a Honeywell technician would show, followed sometime later by a BBN
> person to check it out. I figured I had at least two weeks grace. No
> such luck. They fixed hardware glitch, put the IMP on a plane and
> delivered it two daysa EARLY. Worse yet, they plugged it and it
> worked immediately :)
>
> Steve
>
> Sent from my iPhone
>
>> On Nov 23, 2025, at 6:16 PM, Vint Cerf <vint at google.com> wrote:
>>
>>
>> I made heavy use of the fake hosts for generating traffic, absorbing
>> traffic, reflecting traffic (ping) and for tracing paths taken by
>> packets.
>>
>> v
>>
>>
>> On Sun, Nov 23, 2025 at 5:51 PM Jack Haverty via Internet-history
>> <internet-history at elists.isoc.org> wrote:
>>
>> There was a process involved in adding an IMP to the ARPANET. It no
>> doubt changed over time, so what happened with IMP #1 and #2 may be
>> different. I tried asking Ben Barker, since he is mentioned in
>> that
>> 1969 log entry, but the email addresses I have sadly no longer work.
>>
>> IMPs had "fake hosts" which acted like real hosts but were
>> implemented
>> as software inside the IMP. Back in 2012, Ben was helping with the
>> resurrection of the old IMP code and told me:
>>
>> "The IMP code supported a number of "fake hosts". The TTY was
>> one. DDT was
>> another. There may have been another 1 or 2; I had no use for
>> them if they
>> existed, and do not remember them.
>>
>> Each IMP had a hardware card that contained its IMP number, I
>> believe.
>>
>> There was a bit in the header that specified whether the message
>> was for a
>> real host or a fake host.
>>
>> I believe that the TTY was fake host 1 and the DDT was fake host
>> zero.
>>
>> I believe that the default was that the TTY would be set to
>> connect to the
>> same IMP's DDT."
>>
>>
>> Part of the process of getting an IMP online was to first get the
>> two
>> IMPs talking to each other. The TTY on one IMP could connect to
>> the TTY
>> on the other IMP, so the engineers at the two sites could "talk"
>> over
>> the net. You could also connect to the remote IMP's debugger
>> (DDT) and
>> examine or modify the remote machine's memory.
>>
>> After all that connectivity was working and the IMPs were judged
>> functional, then the actual physical hosts could test their
>> ability to
>> communicate.
>>
>> So it's possible that the fuzziness in the timeline was because
>> it may
>> have taken a few weeks to run tests, get all the bugs out, and then
>> declare the first 2 nodes of the ARPANET, and the NCPs in their
>> attached
>> hosts, to be "in service". The stories I've heard recounted
>> that the
>> initial testing crashed with only the LO part of LOGIN successful.
>>
>> One of the other fake hosts in the early IMPs was the one that sent
>> measurement data to UCLA. More than a decade later, when we were
>> involved in getting TCP/IP deployed, we noticed that the UCLA
>> fake host
>> was still in the code. So we used it to create a "Remote
>> Datascope" for
>> DDN (of course, worked in other ARPANET clones too) that could
>> capture
>> the TCP and IP headers of host-host traffic -- invaluable for
>> figuring
>> out what TCP was doing, especially as newly-written
>> implementations by
>> big government contractors were coming online.
>>
>> I doubt anyone knew, in 1969, what was starting then and how
>> important
>> it would be.
>>
>> /Jack
>>
>> On 11/23/25 13:04, Steve Crocker wrote:
>> > After seeing the back and forth on this thread, I now think 21
>> Nov 1969 was when Larry Roberts came to UCLA. I'll try to check
>> with Kleinrock.
>> >
>> > Steve
>> >
>> > Sent from my iPhone
>> >
>> >> On Nov 23, 2025, at 3:48 PM, Jack Haverty via Internet-history
>> <internet-history at elists.isoc.org> wrote:
>> >>
>> >> Perhaps someone in the Bay Area can go to CHM and ask what
>> actually happened on November 21, 1969...? /Jack
>> >>
>> >>> On 11/23/25 12:22, Barbara Denny via Internet-history wrote:
>> >>> Here is Don Nielson's response to my inquiry. I did cut
>> out the first paragraph as it only was for me. There is a typo
>> regarding the 1972 ICC meeting.
>> >>> barbara
>> >>> ----- Forwarded Message ----- From: Don Nielson
>> <nielsonz at pacbell.net>To: Barbara Denny
>> <b_a_denny at yahoo.com>Sent: Friday, November 21, 2025 at 10:05:40
>> PM PSTSubject: Re: Fw: [ih] First ARPANET Link Put Into Service:
>> November 21, 1969
>> >>> "*** paragraph deleted*****
>> >>> I'm certain about the following:
>> >>> 1. First ARPANET connection - Between Bill Duvall at SRI
>> and Charlie Kline of UCLA
>> >>> on what has been accepted as 29 Oct 1969. I tried my
>> best to confirm the date
>> >>> while this date was being pushed by Len Kleinrock of
>> UCLA. Even I and Marc Weber
>> >>> the CHM dug into Engelbart's dinky handwritten
>> notebooks and other stuff on file at
>> >>> Stanford. As far as we could tell, nothing was noted on
>> the SRI end, indicating no big
>> >>> deal at the time. So, what has come to be accepted
>> derives from a brief jot on Charlie's
>> >>> scratchpad: "22:30 Talked to SRI Host to Host".
>> Charlie and Bill are still around
>> >>> and have confirmed and elaborated on the incident,
>> being quoted in some places.
>> >>> 2. ARPANET demo at the International Computer Communication
>> Conference in WDC
>> >>> of NCP in Oct 19723.
>> >>> 3. First 2-net demo of TCP was on 27 Aug 1976. PRNET and
>> ARPANET.
>> >>> 4. First 3-net demo of TCP was on 22 Nov 1977. PRNET,
>> SATNET, and ARPANET.
>> >>> All this is easily available so 21 Nov 1969 seems an
>> aberration by someone.
>> >>> Take good care, Don
>> >>> On 11/21/25 10:14 AM, Barbara Denny wrote:
>> >>> Hi Don
>> >>> Hope you are doing well. This came up on the internet
>> history list. I am wondering if you could shed some insight on
>> the log.
>> >>> Hope you have a good holiday.
>> >>> barbara
>> >>> ----- Forwarded Message ----- From: Vint Cerf via
>> Internet-history <internet-history at elists.isoc.org> To: Jack
>> Haverty <jack at 3kitty.org> Cc: "internet-history at elists.isoc.org"
>> <internet-history at elists.isoc.org> Sent: Friday, November 21,
>> 2025 at 09:50:33 AM PST Subject: Re: [ih] First ARPANET Link Put
>> Into Service: November 21, 1969
>> >>> crocker and I were wondering the same thing off the list.
>> >>> v
>> >>> On Fri, Nov 21, 2025 at 12:49 PM Jack Haverty via
>> Internet-history <
>> >>> internet-history at elists.isoc.org> wrote:
>> >>> > Can anyone explain why the article says "What Happened
>> on November 21st"
>> >>> > but the image of the "First ARPANET IMP log" shows "29
>> OCT"? /Jack
>> >>> >
>> >>> > On 11/21/25 09:37, vinton cerf via Internet-history wrote:
>> >>> > > there are so many milestone dates....
>> >>> > >
>> >>> > > v
>> >>> > >
>> >>> > >
>> >>> > >
>> >>> > > On Fri, Nov 21, 2025 at 12:03 PM Frantisek Borsik via
>> Internet-history <
>> >>> > > internet-history at elists.isoc.org> wrote:
>> >>> > >
>> >>> > >> Happy birthday to the Internet!
>> >>> > >>
>> >>> > >> https://www.computerhistory.org/tdih/november/21/
>> >>> > >>
>> >>> > >> All the best,
>> >>> > >>
>> >>> > >> Frank
>> >>> > >>
>> >>> > >> Frantisek (Frank) Borsik
>> >>> > >>
>> >>> > >>
>> >>> > >> *In loving memory of Dave Täht: *1965-2025
>> >>> > >>
>> >>> > >> https://libreqos.io/2025/04/01/in-loving-memory-of-dave/
>> >>>
>> >>>
>> >> --
>> >> 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
>> >> <OpenPGP_signature.asc>
>>
>> --
>> 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
>>
>>
>>
-------------- 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/20251123/8c2df23e/attachment.asc>
More information about the Internet-history
mailing list