Join us as we gather around the hedge, where we dig into technology, business, and culture with the finest minds in computer networking.

Well, hello, Jeff.

How are you today?

Hi, Russ.

No.

I'm good.

I'm good.

It's down here in the Southern Hemisphere.

It's You're upside down.

Summer.

Yeah.

I am I am upside down, and things are getting, you know, warm and green, which, I love.

Here, it was.

The high today was around 10 degrees centigrade or a bit less.

And I don't love that.

And overnight, it was, I don't know, down into the twenties here, which is negative eight to negative 10 centigrade something along those lines.

Not feeling the love for me on that.

No.

No.

That's that's pretty actually, pretty common for Tennessee.

This for Eastern Tennessee and being in the foothills of the mountains this time of year.

We're on the West side of the mountains, and so we get all the weather that doesn't make it over the mountains.

It all stops with us because it's So You're just lucky, I guess.

Well, we don't get the nor'easters that come up the East side of the mountains.

And, actually, the nor'easters are horrible.

That's the worst weather in The US, I think, is the nor'easters that go up the East Coast, that come up the Florida Coast, hit North Carolina, South Carolina, and then work their way all the way up the coastline.

Holy mackerel.

Those things are those are some crazy, crazy storms.

But, anyway, that said, we were talking by the way, Jeff, I mean, maybe you wanna say hi to people before Oh, hi, everyone.

Yeah.

I'm Jeff Houston.

As might might be obvious, I I live in the Southern Hemisphere, in Australia to be precise, on the East Coast where right now it's all turning gloriously summer, which is highly enjoyable.

And, yes.

That's me.

I work as the chief scientist at APNIC, which means I get paid to think thoughts.

Oh, cool.

Awesome.

Do.

Think thoughts.

So before we started, we were gonna talk about Moore's law and the Internet and such.

And you started off talking about telephone networks.

Now if you've listened to me for a while, some people don't know this, but I started out in The US Air Force doing electronic.

Actually, I started long before the US Air Force doing electronics.

I got my immateriality license when I was 12.

So I was actually doing 40 words a minute Morse code when I was 12 years old, and I was doing climbing towers and hanging inverted v's and long wire 40 meter and 10 meters and hand adjusting and soldering together YAGI antenna antennas and all those LPDs, log period dipoles, and all that other stuff.

When I was in the air force, we had a an FPS 77, which was a storm detection radar, which was installed in 1964.

It was serial number one.

It was a weather radar, serial number one of the mid range.

There was one earlier.

There was a model before the 77, then they came out with the 77, and now they have Nexrad, which is all computerized stuff stuff.

And this was all tube type.

We even had tube type operational amplifiers, and it was all high voltage.

Everything was high voltage in the whole thing.

Absolutely.

Yeah.

Yeah.

The magnetron was fed by a flywheel circuit that pushed, I, a 100 watts at, you know, one volt or 10 volts or something.

But you said it was all same.

It it was all tubes.

Yeah?

Yes.

Magnetron, receive circuit, everything was all tubes.

There were no digital electronics in it.

And then on the other side of my world I mean, I worked on lots of other things that were like that, very old.

I worked on the telco network from time to time.

And k g 80 fours, which I don't know if you know much about 2 threes and k g 80 fours crypto gear.

But the telco system was based on stroger switches.

We had a 10,000 line frame that was based on paper wrapped cable and stroger switches.

Right.

Right.

And, you know, that was the the early, you know, computer sorry, telephone networks were effectively electromechanical.

The stroger switches actually really created an electronic circuit.

And the issue was, you know, in some ways, technology was extensive.

So all you could do at the other end of the wire out in people's houses and offices was to put a speaker and a microphone.

Yep.

And the power came from, you know, the thing in the middle, And all the switching was in the middle.

And when they first started doing digital switching, then you only had valves.

You only have really expensive power hungry things where you had to share that capacity across thousands of people to make it pay.

And so the telephone network was incredibly network centric.

You know?

Yeah.

It really was stashing all the money in the middle and then sharing that functionality to people at the edge, so highly network centric.

Yeah.

And a lot of the terminology we have today, by the way, comes from this.

So the stroger frame on Maguire, we called it the mainframe because it was the frame that all of the tele telco circuits came into.

Now there were IDFs and BDFs and, you know, subframes and and other things like that, but the frame there that was the stroger was the mainframe.

And so it was called the mainframe.

Right?

And the Right.

It was terminology.

And and, you know, that that was the early design.

But there was a couple of thoughts that were completely off the wall crazy.

And because the telephone network was the concentration or one of the concentration points of technology because if you think about it, you know, there are a few giants around at the time, and one of them was RCA, the broadcaster.

But there was also, you know, dear old Marbell, the telephone company.

And they were desperately trying to make the telephone network bigger and cheaper.

And so their research labs were trying I suppose, I guess, there was a department trying to make valves that were smaller and use less power because, you know, trying to perfect what you have is always a good idea.

But down the end of the corridor, god knows where, but down the end was this strange sort of group trying to push current through this weird stuff of rare earths and and so called semiconductors.

Where they weren't like copper.

They didn't just always conduct.

But it wasn't like stuff that doesn't conduct.

It was sort of in between.

It sort of conducted.

The name goes semiconductors.

And it was in 1947 that a group at Bell Labs came up with an arrangement of silicon, which had been permeated with a certain amount.

I think it was gallium at one point.

I've forgotten, but it doesn't really matter.

And instead of having a valve with power, vacuum, hideous unreliability, you had these strange semiconductor elements, which with a small push of electrons turned from conducting to not conducting or the absence of a small push, which was all a valve was.

A valve was this large current.

And if you put a small amount of voltage on the grid in the middle, it stopped it or it started it.

So a little thing controlled a big thing.

And in the semiconductor world, which they just sort of invented as a transistor, this little current controlled a big current.

Amazing.

Who would have thought?

Which which, by the way, happens with tubes.

This is something like a tube, essentially, a vacuum tube is essentially a cathode and a plate, and then you have a grid in the middle.

And the whole thing about a tube is is you can have multiple grids.

So basically, what you do is you shoot a stream of electrons through the vacuum in the vacuum tube at the plate and through the plate so that the receiver, the cap the to the plate, and the plate picks it up.

And so what you do is you control the strength of the str of the stream of electrons, the amount of voltage that you're getting through the tube, by modifying the amount of voltage you drive on the on the on the grid.

So you can have three or four grids and have them do things.

So this is how you modulate signals, is you actually can adjust.

You have a you have a straight pure sine wave going into the cathode side, and it is if if you don't have any if you don't have any grid, you just get the same straight, you know, out of the the the plate side.

But you put that grid in the middle and you start messing with it, and you can actually modulate signals onto the sine wave.

So it's actually exactly like a transistor, only it's not using these weird silicon things that you you know.

It it it it was the other way.

The the the silicon transistor was exactly like a simple valve.

It wasn't multiple grids and multi unit and so on.

It was really basic and really simple.

Yeah.

But, you know, when they invented it at Bell Labs, the answer from the telephone companies was, well, fascinating, but so what?

Yeah.

Because it wasn't immediately obvious that it had application there.

And interestingly, in the next ten years, the silicon transistor industry went gangbusters in consumer devices.

Transistor radios were a to actually differentiate between that big hulking thing with a whole bunch of valves connected to Maine's power and this little thing that ultimately, you you know, within a few years, could run off a battery.

Yep.

All of a sudden, it was a different world, but it didn't change telephony until and it was in 1958 that Bell Labs came up with something else, which was almost like the multiple grids in a valve in a vacuum tube.

But it was basically multiple transistors on the one substrate of silicon.

Yep.

The integrated circuit.

And then all of a sudden, things turned bananas because we were starting to play with computers.

We played with them a lot in the forties to try and do various feats of of encoding and decoding, various feats around calculations of projectiles and and orbits, etcetera.

We were playing with this stuff, but all of a sudden, integrated circuits meant you could actually build a machine with thousands, tens of thousands of the equivalent of valves.

And instead of taking up buildings, you could make computers that only took up rooms.

Wow.

And at that point, things started to impact back into this wonderful world of telephony.

Because we were starting to uncover the fact that you can actually do this integrated circuits and integrate them with computers.

And you can actually do switching.

Now the switching in telephony, for some reason, the telephone network decided to switch the most difficult thing to switch.

I'm like, you gotta be nuts.

They switch time.

Yeah.

It's kind of, what are you talking about?

That's really hard.

You mean to say, I I sort of create a signal and I pass it through your digital switch.

And out the other end becomes the same signal, the same voice without distortion, without sort of turning into unintelligible.

And, mate, the telephone switch was again a modern miracle, and it was only really possible with with the invention of of the integrated circuit.

And and, again, what was going on was this was still expensive stuff.

Computers were big.

They were kind of custom exotic things.

But there was something else going on.

Because once you'd figured out that what you did to build these integrated circuits was to take away for a pretty pure silicon, I.

E.

Sand, whack a photoresist on it, and then expose it to light and etch patterns into that chip.

And you could apply various impregnated metals to create transistors on on this sort of silicon wafer.

How many?

Woah.

Depends on how good your camera is, to be perfectly frank.

Because two is easy.

A thousand is easy.

10,000?

Well, if you've got a really good piece of optics, a fine lens, and and good positioning, yeah.

You can put tens of thousands of individual circuits on one chip.

And if you do it right, you can do this at a rate of thousands a second.

So at the end of the other side of the production line, sand and a few rare animals get rare minerals go in one side.

At the other end, basically, come to these fantastic sort of combinations of literally thousands of valves all hardwired together to create functionality.

The computing industry was born.

And Yeah.

The first ones were big.

They were really big.

And what's interesting to a lot of people think about the computing side of this, but they don't what they don't think about so much is that without the network stuff that was done early in the days, computers themselves would have been a very limited value.

And there would have probably been 20 or 30 of them in the world at big research institutions that took on specialized tasks.

So a number of things were going on.

And and, you know, computers started to be used in business.

And the whole issue was, I want them cheaper.

I wanna build more of them.

And and and the trick was, can I put more gates, more features on the one piece of silicon?

And the answer was brighter lights, better lenses.

Sure.

Because the process could miniaturize really easily.

And when it miniaturizes, it uses less power per unit of computing.

So over the fifties and sixties sorry.

The sixties and seventies, because the integrated circuit sort of appeared in the world in the late fifties, the sixties, The mainframe computer industry was constantly changing.

They were getting smaller.

They were getting cheaper.

And we started to get not just the mainframe, which was the big, you know, multi tens of millions expenditure, we started to get mini mini mainframes.

Digital equipment were keen on them.

They're called PDPs.

And in the late nineteen sixties AS 400.

AS 400.

Well, PDPs are interesting because, in the same Bell Labs, not the same office, but, Ken Thompson and Dennis Ritchie were mucking around looking at the entrails of a failed project called Multics, which was a grand experiment with MIT and a few others to produce the massive computer that everyone would have an account on, the shared multi user, you know, computing system that would, you know, do everything for everyone all at once.

And they took this PDP 11, tiny machine, a single rack.

That's a 19 inch equipment frame for those who haven't seen one.

They're still around.

They're still the same size.

They took this this PDP 11 and built their own little mini operating system.

The control system that made the computer be able to do things.

You know, it could run programs.

It had a file store.

It had a way you could type in commands.

It had a display.

And and they called it as a play on Multics, UNIX.

So here was this operating system, small, portable, worked on this new generation of of machines that you could afford in a lab, not just afford in a company.

And immediately, when they sort of did this and came to the attention of the the mothership AT and T, there was a problem.

Under the terms of an agreement with the US administration and Department of Justice, AT and T was basically a phone company, and they were a supplier of technology to phone companies.

They built five ESS switches.

They built that technology.

Now they were straying from the reserve if they were doing a business other than equipment for common carriers in the telephone world.

That's not your job.

I went, oh, but we have this really cool operating system.

And they said, not your job.

What can we do with it?

Give it away.

What?

Give it away.

And they did.

So for the cost of shipping, anyone could have a really quite cool lean mean operating system that ran on not just PDP elevens, but because it was written in a high level source code, you can assemble it for almost anything.

And and so by the late sixties, early seventies, this new operating system, UNIX, was certainly a thing.

Every vendor had their own operating system, digital IBM, you know, Burrows, you name it.

But here was this other one, this upstart that wasn't really owned by anyone.

It was open and it was available as we evolved computers and made them smaller and faster.

I want an operating system.

Oh, I'll just put on UNIX.

A morning's work and you're done.

Not five months work of devising a new operating system for this new computer.

It's going, nah.

Just put on UNIX.

I'm out of there.

Revolutionary.

What a thought.

What a thought.

And totally totally accidental.

Totally accidental.

But it made the software independent of the hardware.

So that's the second of these serendipitous thoughts that kind of came from nowhere.

Maybe it was a particularly strong coffee that day.

I don't know.

But, you know, kaboom.

It just happened.

And there was one more, and it was 1973.

Because one of the outcomes of this work in radar and so on was actually using radio for communications.

That's pretty easy.

Now you send a signal, in radio terms, and then you modulate that.

So, So, you know, dot dot dash dash, all that kind of stuff is humming, not humming in radio.

And over in the Hawaiian Islands, they built this radio based network, relatively primitive, pretty slow, called AlohaNet.

And it was kind of neat.

It was a simple system where you waited for a while to see if that carrier frequency was idle.

And then you sent your your little packet of data over to the main hub point.

And then it would send it to wherever you wanted to address it.

It was a neat system because it just used radio.

Nothing else.

All the machines, the computers at the sort of connect to the network formed part of the functionality.

One of the guys who worked on this was Bob Metcalfe, and he was then hired by that well known computing company, Xerox.

Not the the only company that I think that made at least two, maybe three fundamental inventions in computing.

And at the end of this, managed to stay a photocopier company.

Yeah.

So I sit there and think, how did you miss it?

Every time.

Missed.

So this guy, Bob Bittcalf working at Xerox at the time, kind of came up with a wired equivalent of HelloAnet, a wired equivalent.

And this is the beauty of it The network was a wire A wire No electronics Nothing It was the ultimate disappearing trick Because everything was actually in the computers attached to the wire.

And architecturally, it was it was the opposite of telephony.

Really smart concentration of technology in the middle.

It's so expensive.

We can only have one dumb devices at the edge.

This new thing, Ethernet, is the exact opposite.

Computing is now so affordable that every computer can participate in driving the network.

And we can actually take it to the point, and it is a big point to take it to, where the network itself is just a piece of wire.

It's it's nothing.

It's the computers at the edge that are cooperatively talking to each other, driving this thing we call a network.

Wow.

And and when the industry saw this, it's kind of, you mean to say all I need to do is lay wire and I have a network?

And the answer, yes.

Or where do I order this stuff?

And by the mid seventies, this 10 megabit system was being installed almost everywhere.

It wasn't dependent on a single vendor.

You know, Xerox did not make computers.

And oddly enough, that was a good thing because it was openly available.

It wasn't tied up with a particular brand of mainframe.

And and literally, everyone who was looking on building their mini computers and then it got smaller and smaller, Connecting them up was the job of Ethernet.

So we kind of had these three things happening all at once.

Yeah.

Moore's law was starting to take effect.

Your integrated circuits were smaller, cheaper, and faster.

Wow.

We had software divorced from hardware.

And oddly enough, no one knew how to value it.

So we were giving it away.

Just giving it away.

Remarkable.

And then we had networks which were effectively networks that weren't centralized.

They were the ultimate of pushing it so far out to the edge, there was nothing left in the middle.

These were very, very basic inventions, all three.

Yeah.

Yeah.

And now when networks were first done, like, when we talk about mock mock Metcalf days, we actually aren't talking about integrated circuits.

We're talking about discrete electronic circuits, which is a phase of electronic engine of engineering that people seem to miss.

But I've worked on, you know, systems with a hundred, two hundred, 300 gates.

There were literally two there were literally two capacitors and a resistor per gate, all discreetly soldered on to circuit boards.

Because they they I mean, they were transistors, but they really I mean, they were mostly diodes and capacitors.

And Right.

And they were soldered onto physical boards to make gates.

But the trick was you could do it on silicon Yeah.

A lot smaller.

At money level.

And and it was cheaper to produce.

So the cost per gate is a really good way of thinking about this.

Whereas it used to be a couple of cents to actually build the transistor, solder it onto a circuit board, you know, put it with all of its mates, you know, create circuitry that supported the thing working in concert, and then doing the connectors at one end of the the PCB card to connect to all the rest of these, you know, little circuits.

Yeah.

You could shrink all of that to one integrated circuit with a bunch of of pins as its feet.

And and this just kinda took over the world because the cost of production were now down at fractions of a cent per day.

Minions.

And and bizarrely and again, this is something that I think we failed to truly understand.

Moore's law said, every two years as it turned out originally said every every single year, but very regularly, we were able to put twice the number of gates on a single circuit of the same size.

So wow.

This was the incredible shrinking world.

Every two years, let's go two, I could buy state of the art technology that was twice as fast.

Twice as cheap.

And demanded twice as less power than my competitors.

And so the whole reason why bell labs looked at the transistor and said Great.

But we're not ready for it.

Come back and tell us again in a few years' time.

Was that they had a life cycle of equipment of twenty five years, you know, which is why the military, by the way, that that insisted on having that life cycle for decades longer than they should have were using twenty five year old equipment in the seventies because, you know, that was the mode of thinking.

Yeah.

I know.

We were still using nineteen sixties equipment in the nineteen nineties.

Twenty five year lifetime.

And in in one extent, they were right.

You know?

Trans transistors, unlike valves, didn't have a half life of ours.

Yeah.

Transistors had a half life of decades.

Yeah.

And so the thing that you bought in the sixties still worked in the nineties.

But if you're in the competitive world, the commercial world, everyone else was buying equipment and machinery that every two years was twice as good in almost every parameter, the one I have.

So all of a sudden, I've got a problem.

I can't run a twenty five year life cycle on my on my capital, on my equipment, on my asset.

Because unlike the the military and the government, I'm in a competitive world.

I need to change every year or so.

Every year.

And and that really did leave some companies behind and made giants out of the most amazing, almost unintended folk.

Apple was an accident.

Yeah.

It really was.

Because I think very few folk foresaw that when you scale and instead of producing a 100 computers a year, you produce a 100 integrated circuits per second, which turn into what?

Mainframes?

Don't be silly.

Yeah.

How do I sell that extraordinary production capacity?

I need to convince more and more people to buy them.

And so what you do is you get you you release Lotus one two three, and everybody puts one in their closet.

Yeah.

Right.

Yeah.

I build a closet computer, and I convince every single people that are you know, person who owns an apartment or a house, you need one of these in your closet.

And all of a sudden, my market has expanded from a few 100,000 enterprises and businesses games, not by business applications, by the way.

Well, consumers.

Yeah.

Consumers.

Because once you once you enter the consumer market, you know, most of us spend most of our time wanting to be entertained.

And that's that's a fine human pursuit.

I do it myself.

But, you know, what we found then was that these things become a scaling consumer technology.

Mhmm.

Now, again, you look at sort of the the tools that we then had, cheap networks, really, really minimal networking, cheap software.

Some of it was free like like, like a Unix.

Hardware that was ever shrinking.

And we had this kind of desire, if you will, to sign build more of them and link them together.

And and out of that grew the Internet.

And the Internet was kind of the same expression that it wasn't a network that was dense in the middle.

Really wasn't.

Because like Ethernet, it was all at the edge.

Yeah.

All the software that is the Internet, all all the protocols and applications and packages.

I mean, the end to end yeah.

The end to end principle.

Right?

Like, the the intelligence should be at the edge and the network should be dumb.

Why?

Because the network does has a longer life cycle.

It's physical wires in the ground, and so, therefore, you want those physical wires to last for twenty, thirty years.

And, you you know, you do all your upgrades on the edge to make the network more useful over time.

And because Right.

The network itself is a big investment.

Well, boy, have we reversed all of that.

Well, that then leads to the next point.

Because you kinda think, oh, it's just computers, and the stuff in the network's kind of well, it's expensive and fixed, isn't it?

Yeah.

No.

No.

And and we first saw it with the use of the telephone network for dial up modems.

Remember them?

Yeah.

Remember when you used to make the call and these things would start screechy?

You've heard for the first ten seconds or so as it kind of had this screech down the wire.

What was it trying to do by screeching, you know, down the wire?

Well, we were getting smarter because the original idea of putting signals on a wire was to take a carrier tone and turn it off and turn it on again.

The absence of a carrier tone was a zero and its presence was a one.

And in essence, you could put a microphone at the other end and reassemble the bits.

And that was about as sophisticated as it was.

But if you try and go as fast as the carrier tone, you start to think about, well, how do I encode a one zero sequence into a frequency signal that's the same frequency as my bits.

Oh.

So I have to kind of modulate the bits.

I have to kind of set it so that a high voltage is a zero, a low voltage is a one, and somehow manage to to encode that.

Interesting.

It's called amplitude encoding.

It's kind of on off.

Or you can try to be cleverer.

I have four volumes.

No volume at all.

That's a zero.

Full volume.

That's one one.

And there's two intermediate states.

Well, I can encode that as zero one and one zero.

Oh.

Out of the same carrier tone, I've got double the bits.

How clever.

All I need is four amplitude levels.

How about 16?

Well, I can get more.

How about 32?

You know, the human ear is pretty good at this.

It can distinguish about 256 or so discrete levels of volume.

Well, computers can do better because we're using digital signal processes.

And so you start to do really clever encoding.

But is it just all volume?

No.

I can change the phase of the carrier.

Woah.

Big thought.

All of a sudden, by the late eighties, we were getting out of a carrier voice call, which should have been capable of doing 300 bits per second.

By the screeching and the remarkably clever modulation, we were getting 56,000 bits per second.

The carrier tone was actually only eight kilohertz.

8,000 cycles a second.

But we were managing by really clever modulation of that carrier getting seven times the signal out of it.

56 kilobits.

You go, well that's really clever.

You analog guys are amazing.

And then there was a fascinating accident.

Over in Motorola land, they were running out of money.

God knows why.

And they were really clever radio folk.

And they were really clever fiber optic folk.

Really clever.

But as a budget saving measure they put them in the same building.

Now over in the carrier world, we've long since run away from copper conductors for our big circuitry.

We were using fiber optic.

It was, I suppose, cheaper, far more resilient, lasted for years longer than copper, didn't rust.

You know, it was great stuff.

And the technique was you have a torch at one end and the equivalent of an eye at the other end, a digital eye, and you go on or off, On off key And you can control the torch with some circuitry And you can start doing the torch a few million times a second On off on off on off And it kind of works up until a few million But how do you make it go faster?

Well, switching lasers is difficult.

On off keying really does strike a limit.

And and the best they're able to do was just below 10 mill ten ten megahertz, ten minutes per second, which is Ethernet have been doing that for a decade.

It was hardly rocket science in fiber.

I'm just sorry.

I I tell a lie.

I got confused.

They're up to a 10 gigabits per second in on off keying.

That's sort of where it got to.

Right?

Which is okay.

But because they're in the same lab as the modem people, the modem folks said, well, you know, photons, electrons, same thing really, isn't it?

Quantum mechanically, you know, just waves, dude.

It's all waves.

Why don't we phase modulate the light?

Right.

Can you?

Yes.

Sure.

And amplitude modulate it.

And all of a sudden with better digital signal processes, you can push more and more capacity through the same piece of fiber.

Nothing's changed on the glass, but everything's changed electronics at either end.

Everything.

So instead of having a carrier network that ran at a few, well, 100 megabits per second, a few gigabits per second, I'm talking 20 gigs, a 100 gigs, 200 gigs.

And all I did was change the ends.

Yeah.

So yeah.

I mean, it's a bit more complex than that because you're actually in in QSQ or quad quad keying.

QSQ.

Yeah.

Yeah.

You're you're on sidebands.

You're dealing with multiple sidebands.

You know, like, I used to run a radio that was a single sideband, but they were running all eight sidebands.

And there's they were modulating onto eight sidebands or four side bands with more cymbal depth.

So, I mean, there's a lot more going on there.

But, yeah, basically, we, you know, got to the point where we were modulating the same way on fiber as we were on, on copper.

And, you know, there were there were light based modulation schemes before.

They just weren't ever put inside of a cable.

Well, that work that it didn't work cost effectively, Moore's law.

But if you wait long enough, you know, the the the gates get cheaper.

You can put more on a single chip.

I can do more on a single integrated circuit, so we did.

Now I'm gonna swing back a bit a little bit and talk a little bit about the way the telephone networks actually worked.

Because if everyone lifted up their handset at the same time and tried to dial, nothing would happen.

Yep.

There wasn't enough network to go around.

It's they're non blocking, but they're non blocking because you have access control.

You can say no.

You can say no.

You're not allowed to make a phone call.

And you looked at some of the telephone control rooms, and I remember visiting one here in Australia, and they had the TV running on massive screens at the front of the room.

Why?

Because if there was an earthquake in in Italy, some natural disaster in Japan, they would block the calls manually.

I'm sorry.

No one's ringing up Italy right now because, you know, everyone's trying.

It won't fit.

But you think again about what they were trying to do.

And what they were trying to do was ration inadequacy, scarcity.

There wasn't enough to go around.

And indeed, the whole margins of why telephony made us so much money was because there wasn't enough to go around.

So as well as the cost of I run a telephone network, there was the opportunity to charge what the economists call a scarcity premium.

To actually damp demand, you charge a whole lot more than it costs just to dissuade everyone from picking up the handset at the same time.

It's it's the economics of scarcity, and the telephone companies became the became the rationalist and became very rich.

But Moore's law was the insidious worm that was undereating the entire foundation of this.

Because when you applied Moore's law to these digital signal processes and the carrier's own networks, all of a sudden, they didn't have not enough to go around.

They had so much.

It wasn't funny.

It was abundance.

And no matter how they tried, they couldn't put a damper on that abundant capacity.

So all of a sudden, there was this suspicion that the telephone companies were eating everyone.

They they were monopolistic.

They were gouging.

They were they were working against you and I.

And to hang the the era of deregulation, which was actually the era of entering a different phase, the phase of abundance, the utopia of Moore's law.

So now we're designing networks about abundance.

Where transmission is so cheap, it's worthless.

Didn't someone sell Sprint's carriage network to someone else for a dollar?

Yeah.

Pretty much.

Yeah.

It's not cheap.

Much.

They paid way too much because it's not worth anything anymore.

It's abundance.

And you think again about what the job of the Internet was.

It was to take you and your computer over to that server a long, long way away to get the product or the service you wanted.

Dear old Microsoft was having this room full of servers to distribute the Windows operating system in Seattle.

And all of a sudden Seattle was a networking black hole.

It just couldn't stand it.

What about you with abundance?

Well, hang on a second.

Computers are cheap now.

Storage is cheap now.

Processing is cheap now.

Why don't I take what you're trying to distribute and distribute a copy everywhere right beside the user?

The content distribution networks.

And it's all about cost.

It's cheaper to do it that way than to carry the bits.

Even though we've made transmission cheaper, computing got even cheaper.

So did storage.

And so you find companies like Netflix is a good example, Google's YouTube, where they take the entire inventory of what they have, that digital inventory, and reproduce it 10,000 times close to everybody.

Mhmm.

It's the great vanishing network all over again.

Yeah.

Because all of a sudden, the number of miles that packets travel from where they were stored in a storage device to my computer on, you know, in my hand or on my desk, they've shrunk to close to nothing.

Wow.

We've taken distance out of networking.

You see, networking always had the distance square rule.

You know, it comes from physics that force over distance, force reduces, by a factor of distance squared because, you know, that's the way geometry works.

The circumference of a circle.

Sorry.

The area of a circle.

Distance squared.

So cost is distance squared.

Performance is distance squared.

Everything is distance squared.

If you reduce distance, things become phenomenally cheap.

Really so.

So we took distance out of networking.

When we took the networking out of networking.

And all of a sudden, today's networks are remarkable.

We've gone from a provisioning model of, well, it's over there if you need it.

To, look, I bought it to your doorstep along with a thousand, a million other artifacts.

Yeah.

If you want it, just open the door and pick it up.

It's just sitting there waiting for you.

It's just in case delivery.

Just in case you want it, I've already done it for you.

I have pre networked it for you.

It's just waiting there.

And so we've managed to actually achieve the one thing that this industry always said.

And I learned that when I was a child sort of starting to think bigger, faster, cheaper.

Pick two.

I can make it bigger.

I can make it faster.

But if I do so, I can't make it cheaper.

I can make it bigger and cheaper, but it won't be fast.

Now there was always a compromise.

But all of a sudden, bigger, faster, cheaper, well, I'm good.

I'm good.

Well, it's and it still is to some degree.

I mean, because you still have to store the data at a thousand, a million locations.

You still have to like, it still costs something.

And so you could If it costs less than a millionth of what it used to cost Yeah.

It actually costs less.

You can replicate it and still pay less.

I mean, you yeah.

You but you could actually still not store the data and still pay less.

But economically, the trade off is such that it's better not to.

It's better said yeah.

Moore's Law said buy more computers.

They're cheaper.

Keep up with your competitive.

You know?

It's a race.

Come on.

Come on.

Come on.

Change.

Change.

Change.

Get on with it.

And so now I think we're looking at at the next revolution.

And it's interesting because we've lived through Moore's Law, but things are looking hairy.

And this is, I suppose, the the portent, which is a useful place to end this.

The feature size on today's silicon is down around two nanometers.

It's remarkable to get there.

We have to vaporize Indian tin using laser pulsing to get a light signal of nine nanometers in order to focus it down onto a chip to etch features that are two nanometers wide.

Yeah.

Wow.

What do we do next?

Oh, let's make it down to one nanometer.

Woo!

How big is a silicon atom?

Well, a fifth of a nanometer So I get five of them across a one nanometer track Just five Now the whole idea of semiconductors was that electrons are well behaved.

And if I put a slightly excess amount, it can change a conductor to a non conductor.

Right?

But this kind of relies on the polite behavior of electrons.

But when you get down to five atoms, they're not very polite.

And in fact, they're not even electrons.

They're a wave function and they leak.

Can we make gates at one nanometer?

Well, no.

It just can't.

They don't behave well enough.

So at some point, you've got to be a lot cleverer to get the next doubling.

I can't just make it smaller anymore.

We've been making it smaller since 1957.

Wow.

You know?

So we've done sixty five years of making it smaller.

And we've now got to the size of an atom.

And it's kind of, dudes, you're done.

In this this universe, in this model of physics, you know, physics version one, that's it.

You you've got to go elsewhere.

And, you know, you've go into the fuzzy world of quantum computing, and my brain just exploded a second ago because I don't understand it.

Or you can try and build three d silicon lattices at two nanometer feature size.

And and quite frankly, you know, that's a challenge.

In anybody's language, that's a challenge.

Yeah.

Yeah.

Or you can say, have we reached the end of this road?

And and it's sort of a plausible question.

And and that's what I'd like to leave you with, because if you look at the feature size and processing capacity of cores lately, instead of making it, if you will, smaller and smaller and smaller, what we've actually been doing is adding to the number of cores at each circuit.

So it's integrated circuit process.

It now has 256 cores, but it probably won't have 512.

And and even if it does but even if it does, you're still looking at now a chip die that is Bigger.

One and a half to two times larger that consumes one and a half to two times the amount of power.

So you're really not gaining a lot.

And it has 10 times the failure rate in production Yeah.

Etcetera.

So all of a sudden, this is this next generation big chip is actually more expensive than today's generation small chip.

And and it's kind of what have you won?

And the other thing is too, the physically bigger chip, it takes more time to get from one side to another.

Yeah.

Again, it's slower.

So we're topping out.

CPU processing, three gigahertz clocks, not really getting faster in production.

Feature size, we might hit a trillion features per chip.

But beyond that, there's no clear indicator.

Because unless someone thinks of something as out of the box, so much as the transistor and the integrated circuit, something else, Unless something else walks in the room, we've kind of milked this for all it's worth.

And and my closing thought is, well, what does that mean?

This means that the future is no longer a competition to the current.

I can go back to twenty five year life cycles on my equipment.

Yep.

I can be as big as I want without worrying about my competitors.

And this is even happening down to the cell phone market and even to the laptop and the the home computer market.

Right?

Like, okay.

Used to be I replaced my computer every two years.

Why?

Because as I upgraded software, it gets slower.

Yep.

Now It's it's well, no.

You have to upgrade.

No.

You have to upgrade because the current model comes in black.

Yes.

That's that's exactly right.

That's what's happening.

Next year next year's will come in gray.

Yeah.

You know?

And that's the the Yeah.

We've moved from capability yeah.

We've come from capability to fashion.

And whenever you get to fashion, like Well, the people who are big stay big, and the people who are small can't compete because the technological edge that gave small its sort of cut through that actually made Google.

Let's index all the worlds of information.

The incumbents at the time were going, laugh point, giggle hysterically, you're nuts.

Now Google's the behemoth because it did it.

And we're gonna strike this right now as a closing thought about AI.

Because the current thought about AI is, as long as it gets cheaper and bigger and cheaper and bigger, this will work.

It doesn't work at the moment.

It's funded by venture capital.

It'll work tomorrow.

Trust me, Moore's law.

But what if Moore's law isn't?

What if it doesn't?

Yeah.

How long does the venture capital last?

Yeah.

People say, just remember about AI that it's the worst it'll ever be.

And you know what?

You should remember the opposite of that, that it may not actually ever get much better.

Well, you're relying you're relying on a universe whose physics are not the physics you live in.

Yeah.

You're relying on something else that we don't understand.

So Yeah.

It's an interesting thought that we've lived out sixty five years of phenomenal, prodigious growth coming from those subtle inventions of the transistor and the integrated circuit.

And we've milked them for all they're worth.

And and the intrusive little thought is, is that all there is?

Have we milked it as far as it will go?

And if the answer is yes, start to think differently about tomorrow's world.

Yeah.

It might look a lot more like the nineteen twenties, the nineteen thirties.

Thereafter, there was a huge burst of innovation in the mechanical world, and then everything kinda settled down for a couple of decades and, you know.

Yes.

It's it's the same kind of of of issues flying around here.

And and, you know, that's the transition we're going through at the moment.

So, I leave I leave our listeners with with that thought that, you know, this particular oh, it's a happy thought.

Oh, but the opportunities of tomorrow, we're not just making today smaller.

Tomorrow will not be like, you know, will be it's a different kind of world when you just can't rely on the prodigious nature of Moore's Law to fix everything up, including AI.

You've gotta do something else with it.

Yeah.

Yeah.

Exactly.

Well, thanks, Jeff, for that, long or short or however you wanna put it, that trip through history.

And, I don't know.

I don't have much else to add to that.

So do you have a place where people can follow you or get in touch with you if they want to?

All my ramblings are at dubdubdub.potaroo.net.

You might need a spare hour or two, but I promise I try and make it varied and entertaining with the kind of stuff I put up there.

But they ain't tweets.

Oh my god.

No.

They're not tweets.

I'm sorry.

This is actually the curse of the writer.

Even when I write short post on LinkedIn, they're, like, 300 words.

And, like, I just can't I just I just I'm sorry.

I can't.

Right.

So, you know, if you want the slightly condensed version, you've got it in one hour on this podcast here and there.

That's as short as it gets.

Exactly.

So I try.

Alright.

Cool.

And I'm Russ White.

You can always find me here at the hedge at rule eleven dot tech on x occasionally, whenever I feel like logging into that account, and LinkedIn.

Thanks for listening to this episode of The Hedge.

We know your attention is we live in an attention driven economy, and your attention is high value at this point.

So we thank you for listening all the way to this bitter end of the hedge, and we will catch you next time.