[ih] Nit-picking an origin story (touch at strayalpha.com)

[Jack Haverty]

When I got to Lick's group at MIT, the IMP was already installed. Bob 
Metcalfe was finishing up the hardware interface for our PDP-10, and Bob 
Bressler was working on putting NCP into the ITS OS.  I don't recall 
ever seeing the actual circuit, but the connectors on that IMP were 
quite impressive chunks of metal.  Looked quite suitable for hooking up 
to a "firehose" of data.

I never had any involvement with the actual physical IMP circuit, until 
1972 when I was one of the crew setting up for the ARPANET demo in the 
Washington Hilton.  IIRC, as a 20-something less than a year after my 
degree, it was my first actual "business trip".

By sheer chance, the tech from The Phone Company encountered me first as 
he unrolled a hank of wire that disappeared into the wall. He was there 
to hook up the circuit to the IMP (actually TIP) so it could connect to 
the rest of the ARPANET.  He asked me something like "Where do you want 
this?".  I led him to the TIP in the center of the raised floor we had 
just installed, and helped get the wire underneath the floor.

I was greatly underimpressed.  The "firehose" of 50 kilobits per second 
was just a few wires (4 IIRC) covered by some kind of fabric 
insulation.   It looked more like the kind of wire you used to hook up a 
doorbell.  Not what I expected to go with those massive connectors on 
the IMP.

Jack



On 8/24/25 18:01, John Shoch via Internet-history wrote:
> As I finished writing this I just saw Guy Almes' comment on 50 Kbps analog
> modems vs 56 Kbps digital lines -- thanks, Guy.
>
> Allow me to offer a bit more context about how (I think) this evolved --
> and try to separate some apples from oranges.
>
> When Tom Marill of CCA first explored a cross-country data experiment in
> 1965 he listed some telecomm choices available at the time (although the
> discussion mixes in issues of technical matters and tariff considerations):
> --AT&T Direct Distance Dialing (DDD).  Dialed phone calls via the analog
> voice network, with an AT&T "data set" (modem), typically 2-wire with
> per-minute charges.  A 103A data set could get 300 bps.
> --Wide Area Telephone Service (WATS).  This was really a tariff overlay,
> allowing discount levels for high use, with some tariffs for unlimited use.
> --Western Union Broadband Exchange:  This was a dialup service offered by
> Western Union built upon their telegraph/Telex infrastructure, NOT on the
> AT&T long distance network.  It was meant for data use, had 4-wire
> circuits, but still needed a modem.  The other party had to have a WU
> connection, as well.
> --There were also private, dedicated, leased lines available from the phone
> companies.  These were a lot more expensive.  You still needed a modem, and
> a tariff plan.  He mentions the Telpak tariffs which allowed you to bond
> together, for example, 12 voice channels for a total of 50 Kbps.
> --He also notes the need for auto-dial and auto-answer devices, to allow
> unattended dialup connections.
>
> By the time of the Marill and Roberts paper at the 1966 FJCC they describe
> the experiment they propose to undertake with dial-up, on demand, serial
> links providing terminal access between 2 machines:  ""Initially, a 4KC
> four-wire dial-up system will be used with 1200-bit-per-second asynchronous
> modems."  The reference to four-wire access is a good hint that they will
> not try to use AT&T dial-up lines.
>
> In a (final?) report several years later CCA described what they actually
> did:  dial-up access from a user on the MIT TX-2 to the SDC Q-32:
> --They connected to the Western Union Broadband Exchange service with
> dial-up access.  Auto-dialer at MIT, auto-answer at SDC.  (SDC did not
> install an auto-dialer, and thus could not initiate a call to MIT.)
> --4KC analog channel, data sets at each end, tariff of $.75 per minute.
> They report using 1200 bps Western Union 2121B modems (but I can't find any
> information on those units).
> --It provided terminal access for a single TX-2 user to the Q-32 time
> sharing monitor (no addressing to machines nor to processes at the Q-32).
> --In terms of actual usage they reported over an 8 week period:  94 calls
> attempted with 78 calls completed successfully. Assuming 5 work days/week,
> that's about 2 terminal sessions per day.  Average time for call set-up on
> successfully dialed calls was almost 20 seconds,
> [An editorial note:  AT&T was selling lots of modems for
> terminal-to-computer access, and this was a great demonstration of remote,
> serial, terminal-to-computer-to-computer access from one time-sharing
> system to another.  It has been asserted that they "...connected the TX-2
> computer in Mass. to the Q-32 in California with a low speed dial-up
> telephone line creating the first (however small) wide-area computer
> network ever built. The result of this experiment was the realization that
> the time-shared computers could work well together, running programs and
> retrieving data as necessary on the remote machine...."  When you read the
> final report, though, I find that description of a "wide-area computer
> network" a bit...generous.]
>
> During this period Roberts was working on the ideas for the Arpanet.  As
> well reported, it's Scantelbury (from the NPL in the UK) who urges him to
> seek out higher-speed connections.  As I recall, there are reports that
> Roberts realized that the government had access to better tariffs for use
> of leased lines from AT&T.  Telpak provided a tariff for 12 analog voice
> channels bonded together to provide a total of 50 kbps.
> AT&T describes the use of the 303 wide-band data station/modem (sort of the
> size of a 2-drawer file cabinet):  "The next lower convenient breakdown is
> the "group" channel which uses the bandwidth of 12 voice circuits. The
> 303-type equipment can transmit at a synchronous speed of 50 kilobits per
> second over group facilities." 
> https://bitsavers.trailing-edge.com/communications/westernElectric/modems/303_Wideband_Data_Stations_Technical_Reference_Aug66.pdf 
> The BBN papers on the Arpanet design at the 1970 FJCC, by Heart, 
> Ornstein, et al., describe the 50 kbps links between the IMPS 
> (presumably through the AT&T modems). "Implementation of the subnet involves two major technical activities:
> providing 50-kilobit common carrier circuits and the associated modems...."
> That's 50 kbps through multiple analog channels -- and the basis for
> the "50 kbps" line rate for the Arpanet.
>
> This was all followed later, in the 1970's-80's, with more direct access to
> the underlying digital hierarchy in the US.  Recall that in the US a T1
> digital circuit carries multiple voice channels with a total capacity of
> 1.544 Mbps.  The smallest unit is carrying one digitized voice channel --
> 8K sampling at 8 bits/sample = 64 Kbps for a DS0 link.  To directly carry
> data, though, they have to steal some bits for overhead -- taking out 1
> bit/sample the available throughput is 8K sampling at 7 bits/sample = 56
> Kbps.  Not that this is an entirely digital link, requiring a special
> DSU/CSU at each end (not an analog modem).
>
> Of course, modem technology continued to advance.  By the 1990's (?) they
> had reached about the limit for what you could cram into an analog voice
> channel -- a 56 Kbps modem.  (As I recall, though this was not 56K in both
> directions.). Note that this goes from digital in the computer, to analog
> out of the modem and into the dial-up line, to digital for carrying through
> the digital voice network, back to analog, out to the receiving modem, and
> back to digital for the computer.  Phew!
>
> John Shoch
>
> .

-------------- 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/20250824/8d4a7b8e/attachment.asc>