[ih] How Plato Influenced the Internet

[Steve Crocker]

Thanks for the color.  Lots of things were competing for attention.

I worked on the original host-host protocol, later renamed NCP.  One of the
things I was very concerned about was how many round trips it would take to
establish a connection.  I wanted it to be as few as possible in order to
make the system as responsive as possible.  In retrospect, I wish I had
considered voice and related applications in mind and not expected all
applications to be built on top of a reliable stream of bits or bytes.

Steve


On Wed, Aug 25, 2021 at 3:13 PM Jack Haverty <jack at 3kitty.org> wrote:

> Hi Steve,
>
> I wasn't worrying about efficiency; I just wanted to see if I could figure
> out how to make some ternary logic.   IIRC, the class was on digital
> design, and the lab assignment was basically "build something digital,
> other than the examples we used in the lectures".   Rather than building
> boring flipflops and gates, maybe an adder, I decided to try to create
> ternary logic.   After all, they said "digital", so it didn't have to be
> binary.   I always tended to think outside the box.
>
> Your math is probably correct, but it's only a small part of making a
> design choice.   At about the same time of that lab course, I had a student
> job using a PDP-8 to gather and analyze data from experiments being done at
> the "Instrumentation Lab".   They were designing, building, testing, and
> deploying inertial navigation units that were use in a lot of places,
> including the Apollo spacecraft.  So I actually got to work with real
> "rocket scientists".  I learned as much from those engineers and scientists
> as I did from the classes.
>
> One thing I learned was that much of the mathematical toolbox from the
> courses wasn't terribly useful.  For example, there were lots of tools and
> techniques for minimizing Boolean logic.   But reams of data had shown that
> the critical issue for reliability (it's hard to fix things in space) was
> the mechanical connections involved, e.g., how many pins on a PC board and
> corresponding socket were needed in the system design.   More logic
> circuitry was OK if it meant fewer pins were needed.   None of the tools
> and techniques taught in courses even mentioned the issue of pins or other
> such design questions, such as heat dissipation.
>
> So, it's possible that some kind of non-binary logic might have required
> fewer pins and resulted in more reliable hardware.  Or not.  The choice
> made early on meant we'd never pursue any other path.
>
> Getting back to Plato and the Internet, I can confirm that I never saw or
> even heard of Plato during the 60s/70s/80s.  So it probably didn't
> influence me, at least not directly.
>
> However, I was surprised to just read how Plato was focussed on latency as
> a key driver of the users' experience.   I ran into that same issue at
> Licklider's MIT lab, when we were trying to bring up MazeWars on our
> newfangled Imlacs that were used as terminals on the PDP10.   I spent a bit
> of time tweaking the RS232 TTY interfaces to get the line speed up around
> 100 kb/sec (typical max was 9.6 in those days), and that made the Maze game
> popular.   When you "shot" an opponent they died as they should.  With
> higher latency, they'd often inexplicably get away.
>
> We tried to convince BBN to upgrade TIPs to run faster, but were
> rebuffed.  The TIPs supported the "maximum reasonable speed" of 9.6.
> Nothing faster was needed.
>
> Later on, circa 1978 while we were rearchitecting TCP to split out TCP and
> IP, and introduce UDP, I remembered my experience with latency, and pushed
> for inclusion of "Type Of Service" so that the underlying IP transport
> might someday be able to offer both low-latency and high-throughput
> services to meet different users' needs.   And maybe a "guaranteed
> bandwidth" service to better mimic old physical circuits.
>
> Low latency was also important for things like conversational voice, so
> the "voice guys" at places like ISI and Lincoln were also interested in
> having such a capability in the Internet.   I don't recall that "Plato
> guys" were involved, but I bet they would have been proponents as well.
>
> Sadly, although those experiences certainly "influenced the Internet" to
> the extent that various header fields and rudimentary mechanisms were
> included in the emerging TCP that we still have today, there apparently
> wasn't enough pressure and interest to cause low-latency service to
> actually get implemented.   At least as far as I can tell....
>
> I can't see "inside" the Internet now, just as a user today.   But simply
> watching the now constant stream of live interviews on TV, and the
> pixelization, breaking audio, and such artifacts, makes me conclude that
> low-latency service isn't there yet, after 40 years of evolution.
>
> I suspect part of the cause was also the hardware availability, or lack
> thereof.  Like Plato, Imlacs were not common in "the network community" and
> neither were voice-capable terminals.  So unless you had one of those, you
> didn't understand why things like low-latency were needed.   So the "rough
> consensus" never emerged for such mechanisms.
>
> Plato, and Maze, and Conversational Voice, and no doubt others, influenced
> the Internet.   But not enough to drive the associated functionality all
> the way to deployment.
>
> Sometimes history is about what didn't happen.   And why.
>
> /Jack
>
>
>
>
> On 8/23/21 7:47 PM, Steve Crocker wrote:
>
> Jack,
>
> A classic analysis of bits vs trits says trits are slightly more
> efficient.  The analysis is based on assuming that it takes b parts to
> represent a digit in base b.  Two parts for a bit, three parts for a trit,
> four parts for a quit(?!).  The information content of k digits in base b
> is b^k.  The "cost" is b*k.  The optimal base is e (2.71828...). Bases 2
> and 4 are equal.  (The information content of 2k bits is 2^(2k).  The
> information content of k quits is 4^k.  The costs are the same, i.e. 4k.)
>
> Trinary is better but not by much.  Using six parts, you can make three
> bits or two trits.  The information content of three bits is 8.  The
> information content of two trits is 9.
>
> A different consideration of using trinary vs binary is the representation
> of integers.  As we all know from hard experience, twos complement
> representation of signed integers gives you an asymmetry with one more
> negative number than positive number.  Switch to ones complement and you
> wind up with two representations of zero.  Trinary gives you a naturally
> symmetric representation of signed integers.
>
> I think that's the end of the advantages of trinary over binary.  But I'm
> VERY impressed you took the time and effort to actually build such
> circuits.  Bravo!
>
> Steve
>
>
> On Mon, Aug 23, 2021 at 10:29 PM Jack Haverty via Internet-history <
> internet-history at elists.isoc.org> wrote:
>
>> Back in the 60s, a lot of computer technology was not yet cast in
>> concrete.  There were lots of choices.  But then someone pursues one
>> choice, and if it works reasonably well, others follow the same path.
>> It doesn't take very long for the "installed base" to become so large
>> that it's unlikely that some other initial choice could easily take
>> over.    Think about how long it's taken, so far, for IPV6 to supplant
>> IPV4.
>>
>> Sometime around 1968, as a learning experience in some lab course at
>> MIT, I decided to make some non-binary logic.   At the time, analog
>> computers were still around, and digital computers hadn't yet agreed
>> even on how many bits were in a byte, or how to encode characters, or
>> what order bits should be in a computer memory word.   But bits were
>> pretty well established.
>>
>> I figured there must be other choices.   So I made some ternary logic.
>> Unlike binary, which dealt with 1s and 0s, I used +1, 0, and -1 as the
>> three possible states.  Electronically it translated into positive,
>> negative, or no current.   Using transistors and such components, I made
>> some basic logic "gates" that operated using three states instead of
>> two.  Was that a good idea?   Probably not, but it was a good way to
>> learn about circuits.    Instead of bits (binary digits), how about
>> manipulating trits (trinary digits)? There's nothing magic about 1s and
>> 0s.
>>
>> Shortly thereafter, binary took over as circuitry went into integrated
>> circuits and a whole industry came in to being around binary
>> computers.   If some other kind of approach, ternary, quaternary, or
>> whatever is better than binary, we'll probably never know.   I suspect
>> something might happen soon with qubits though to challenge bits
>> supremacy..
>>
>> There are lots of ways to do things, and the one that "wins" might not
>> have been the best choice.
>>
>> Imagine how networking and computing might have evolved with trits
>> instead of bits....
>>
>> /Jack
>>
>>
>> On 8/23/21 12:15 PM, John Day via Internet-history wrote:
>> > Agreed.  There are only so many ways to do something.  ;-)
>> >
>> >> On Aug 23, 2021, at 14:46, Craig Partridge <craig at tereschau.net>
>> wrote:
>> >>
>> >>
>> >>
>> >> On Mon, Aug 23, 2021 at 8:12 AM John Day via Internet-history <
>> internet-history at elists.isoc.org <mailto:internet-history at elists.isoc.org>>
>> wrote:
>> >> It is not uncommon in the history of technology (it has been observed
>> back several centuries) that it isn’t so much direct transfer of technology
>> but more someone brings back a story along the lines of, ‘I saw this thing
>> that did thus and so and kind of looks like t.’ Which gives someone the
>> idea, that if it exists, then how it must work like this.’ It isn’t quite
>> independent invention, but it isn’t quite direct influence either.
>> >>
>> >>
>> >>
>> >> Related comment -- from my various interactions with historians about
>> technology history.  If the available technology is limited (as it was in
>> the 1950s/60s/70s and early 1980s in many dimensions) then your solutions
>> to certain problems are going to look rather similar.  That doesn't meant
>> that two similar solutions influenced each other... The trick in writing
>> tech history is figuring out where there was a choice space and where there
>> wasn't (much of) one.
>> >>
>> >> Craig
>> >>
>> >>
>> >> --
>> >> *****
>> >> Craig Partridge's email account for professional society activities
>> and mailing lists.
>>
>>
>> --
>> Internet-history mailing list
>> Internet-history at elists.isoc.org
>> https://elists.isoc.org/mailman/listinfo/internet-history
>>
>
>