19

I've heard that the first time a message was sent through it was "lo" since the SDS 940 crashed before the full message of "login" was able to be sent through.

However, I have been unable to find how fast a message was transmitted through it. Does anyone know how long that message must have been taken to transmit, or an estimate at least?

jay
  • 293
  • 1
  • 5
  • 3
    Estimate: The speed was probably bounded by the speed of the leased line, and early modem rates were 50, 75 and 110. 300 baud was after 1971 to my knowledge. Also note that "login" was probably typed, so the actual speed of the characters was human typing speed. As for latency, you probably have two add 2x imp latency + 2x host latency + line latency. IMP specs should allow a guess. – dirkt Jan 19 '24 at 15:14
  • 7
    From an article: AT&T supplied the 50kbps connection between the two sites, with engineers set up an analogue phone call to confirm the receipt of messages. Ref: Who supplied the technology : ARPANET anniversary: The internet’s first transmission was sent 50 years ago today – Brian Jan 19 '24 at 15:18
  • Really cool, I was expecting <30kbps. Faster than I had anticipated. – jay Jan 19 '24 at 15:44
  • 2
    The fact that only 'lo' of 'login' was sent is a reflection on typing speed rather than line speed. Presumably the protocol was character-interactive. – dave Jan 19 '24 at 15:44
  • @another-dave sure, as it was intended as (dumb) terminal session. – Raffzahn Jan 19 '24 at 15:48
  • 5
    And probably sent from a teletype, which requires a rest between keystrokes to regain the strength in your fingers :-) – dave Jan 19 '24 at 16:00
  • "How long is a piece of string?" – tofro Jan 19 '24 at 22:37
  • 1
    @another-dave: It didn't crash because of typing speed. It crashed because the echo reply to "g" wasn't "g" but "gin" and the remote terminal code couldn't handle it yet. – Joshua Jan 20 '24 at 01:04
  • @Joshua I certainly didn't intend to imply that speed had anything to do with crashing; but speed is a major part of the OP's question. – dave Jan 20 '24 at 02:00
  • I seem to remember reading an original TCP paper which had a graph of throughput vs latency measured in minutes. – user253751 Jan 20 '24 at 10:23
  • 1
    @jay note that in the pre-digital-telephone days, phone connections had better quality. The actual quality you got would depend on what equipment and how much wiring was between you and the other person. Basically, local calls set up an actual bona fide electric circuit directly between you and the person you called. Long-distance calls were likely to go through electronic (vacuum tube) frequency multiplexers, which put a hard limit on the signal quality. – user253751 Jan 20 '24 at 10:25
  • 1
    This story sounds apocryphal to me. The original TELNET protocol was half-duplex and line-at-a-time with local echoing. – Barmar Jan 20 '24 at 16:04
  • 1
    The "login" was a prompt, and would have had nothing to do with interactive typing. – jimc Jan 20 '24 at 20:29
  • 3
    @jimc - not according to Kleinrock: 'At the UCLA end, Charley typed in the “l” and asked SRI “did you get the l?” “Got the l” came the voice reply. He typed in the “o,” “Did you get the o?”'. SRI was running an SDS 940 timesharing system, which if I recall correctly had command recognition: type el, oh, and it outputs gee, eye, en. – dave Jan 21 '24 at 19:22
  • There is a possibility apocryphal story about a researcher who was live-patching the system he was working with, and in order to do so at odd hours persuaded the school to install a data link directly to his house. The canonical explanation was that he wanted to be able to perform his feats of daring without a net under him. – keshlam Jan 22 '24 at 03:13

3 Answers3

19

From the IEEE Annals of the History of Computing 44(2) 8-19 (2022) titled Seeking High IMP Reliability in Maintenance of the 1970s ARPAnet on finds

The design called for a small computer called an Interface Message Processor (IMP) to be collocated with each of the computers (Hosts) that ARPA desired to have access to the network. Each IMP was to be connected to 2 or more other IMPs by 50 kbps telephone circuits to form a loose mesh network.

So at least the (leased dedicated) telephone circuits were 50 kbps.

Jon Custer
  • 1,328
  • 1
  • 12
  • 19
4

Does anyone know how long that message must have been taken to transmit, or an estimate at least?

It's two characters, so even considering an inefficient protocol and slow lines (*1) will be considerably less than a second - except it maybe rather be limited by typing speed, wouldn't it?

*1 - Slow by today's speed, as it can be assumed that those were leased lines running at least at 19200 bps.

Raffzahn
  • 222,541
  • 22
  • 631
  • 918
3

When using a packet-switched protocol such as TCP/IP with John Nagle's flow-control algorithm, a system that has sent out a packet of data will hold off on sending further data for the channel until it has received an acknowledgment. Once the acknowledgment arrives, the system will send a packet containing whatever data became available for transmission while the system was waiting for the acknowledgment, subject to packet-size limitations. This will yield "smooth" behavior in situations where turnaround latency is short, but avoid transmitting an excessive number of short packets when latency is longer.

Even on the faster Internet connections that were available in the 1980s and 1990s, it was common to see a noticeable hiccup between the first handful of characters that was sent on a link, and subsequent characters. If something went wrong between the first packet that was transmitted, and the successful receipt of an acknowledgment for it, it would hardly be unexpected that only "lo" would get transmitted no matter how fast or slow the link was.

supercat
  • 35,993
  • 3
  • 63
  • 159
  • 5
    I think you're on the right track - those characters typed in to telnet (or whatever the early equivalent was) were most likely sent in separate packets, due to the timing of the keystrokes, using the earliest TCP/IP protocol. But mentioning Nagle's algorithm and its behavior is a red herring as that wasn't invented at the time. – davidbak Jan 19 '24 at 19:17
  • @davidbak: Fair point about Nagel's algorithm not having been formalized yet, but I'm pretty sure there would have been some mechanism for aggregating bytes for transmission rather than sending a new packet for every byte. – supercat Jan 19 '24 at 19:41
  • Well in those early days I have no idea how sophisticated it was but there was probably a timing component - wait nnn ms and then send what you've got. Unless the early protocol they were using was line oriented... – davidbak Jan 19 '24 at 20:37
  • This was long before TCP/IP, which came into use on the Arpanet on 1983-01-01. The end-to-end protocol was NCP (which was not called that at the time). – dave Jan 19 '24 at 23:41
  • 2
    The host-to-host protocol described in RFC 11, which predates even NCP, appears likely to have been what was in use at the time. See 2.2.2 (b) for login. – dave Jan 19 '24 at 23:57
  • 1
    If you care to dig deeper, our user @LarsBrinkhoff has a github repo with the original BBN tech reports. – dave Jan 20 '24 at 00:04