Vinton G. Cerf
"Father of the Internet"
Sr. VP for Internet
Architecture and Technology, Worldcom
"The Internet Tidal Wave"
Sept. 25, 2000 

Vinton Cerf: I am always looking for some way to encapsulate why it is that the Industrial Revolution and the Internet Revolution have so much in common. They both are going to have, have had and will have, enormous impacts on our social and economic structures. And the best analogy I’ve been able to come up with so far is to compare the Internet with the power generation and distribution network, the electrical network, and electrical motors to computers. Let me remind you a little bit about what you will remember I’m sure from your high school classes in the Industrial Revolution. You’ll remember that when it started, the whole idea was to improve muscle power, human or animal muscle power, magnifying it using mechanical means. And the way we started out doing that was to go find a fast flowing stream. We would build a water wheel and put that in the stream. And we took the rotating energy of the water wheel and turned that into other kinds of mechanical energy using belts and pulleys. The only way that worked of course was to build a fairly large facility right next to a fast moving stream. Then came the steam engine and the steam engine allowed us to move those manufacturing plants to some place other than the fast moving stream, but we still needed water to generate steam. And so typically those steam engines were very large and the plant was similarly rather large and centralized. But the next thing that came along was electricity and we learned to generate it and distribute and we built electrical motors. And the interesting thing about the electrical motors is that they could be built in almost any size. You could build tiny ones, or very, very big ones to do varying amounts of mechanical work. And you could put them anywhere you needed to as long as you could get electrical power to the target. Well I think of the Internet as delivering information instead of electrons, and I think of the computers of the Net doing work on that information the way electrical motors do work with electricity. And in some sense if that analogy is even close to being accurate it explains why the Information Revolution may turn out to be as powerful, maybe even more powerful than the Industrial Revolution that preceded it.

Now I know you all know what the Internet is, but there’s a little reminder here. We speak of "The Internet" as if it’s one net, and in fact it isn’t of course. It’s hundreds of thousands of networks that are all interconnected to each other all around the world. The fact that it is many different and independently operated networks is actually very important for a variety of different reasons; some of which have to do with the economics behind the network; some of which have to do with the way in which it functions and the quality of service that you can expect from the public system. But in this particular instance I want to draw one other fact out: with that many networks, all independently implemented and operated, the only reason this works at all is that they are all using a common set of communication protocols, which we call TCPIP for the name of the two core protocols that were developed way back in the 1970’s. One thing to know about the Internet protocol is that it was designed by Bob Com and me to be as simple as possible. Our theory was that we did not know, way back in the 1970’s, what new telecommunications technologies would come along after we put these protocols and we wanted it to still work. We wanted to future-proof the design. So we picked the simplest possible requirements we could think of. All we asked the underlying communication system to do was to deliver a bag of bits and deliver them from point A to point B with some probability greater than zero. That’s all we asked of the underlying system, and we thought almost any communication system ought to be able to do that. Of course that’s why I think we’ve been successful in putting the Internet protocol on top of literally everything that’s come along is the last 25 years. A-synchronis transfer mode and frame relay and sonic and satellite links and all the other things that have come along including cell phones and third generation communications and blue tooth and all these other radio based systems… MMDS and so on. DSL. I mean you could go on all night with all the alphabet soup. The important thing is that this simple protocol can be put on top of almost anything. In fact some of you have probably seen the t-shirt I had made to commemorate that. It says, "I P On Everything." And that’s basically what I’ve been doing for the past 25 years. Now I don’t have a picture of a new t-shirt I had made though because there’s an implication behind this notion of putting the Internet protocol on top of everything. We have in fact done that. The protocol is now spreading rapidly. But what happens when you standardize things is that you create a platform on top of which you can now build other products and services, and so once the Internet protocol is going everywhere, everyone want s to put their new applications on top of it. And so, in fact, I have another new t-shirt that I didn’t bring today but it says basically not "I P On Everything" but "Everything on I P." Or maybe another way of saying it is "I P Under Everything" because in fact we’re putting all the new applications and protocols on top of this Internet protocol substraight. So that’s the story of the t-shirt.

Now most of you probably already know this, but I found this analogy to be quite helpful to make people realize exactly how simple the Internet really is. If you just think of Internet packets as postcards, you actually know almost everything you need to know about the behavior of an Internet packet because it emulates the behavior of a postcard, except for the fact that the Internet packets go about a hundred million times faster than postcards do. But if you think about it, the Internet packet has a "from" and a "to" address, just like a postcard. And it has some content, and it’s limited content, just like a postcard has only a limited amount of space to put information on it. Now let’s talk about what you know about postcards. One thing you know is that if you put a postcard into the postal service, there is no guarantee that it comes out the other side. This is true of Internet packets too. There is no guarantee that it will actually get delivered through the network. If you put two postcards into the postal service, there is no guarantee that they will come out in the same order you put them in. This is also true of Internet packets. There is no guarantee that ordering will be preserved at that level of the protocol. In fact, there is something that the Internet does that the postal service doesn’t do, as far as I know – if you put a postcard into the postal service, if it comes out at all there is only one. With the Internet it is actually possible to get duplicate packets coming out because of the re-transmission protocols that are used. So that’s even more of a difficult thing to have to cope with. So generally speaking, everything you know about postcards applies to the way the Internet behaves. In fact, if you think about the way the postcards move through the Internet, or the Internet packets move through them, it’s kind of like bucket brigades of computers, special ones called routers, move these packets around. And I think of them as just like a bucket brigade of computers that are handing buckets full of packets from one computer to another until they get to the far end where they get delivered to the destination.

Well that leads to the following interesting question – If this stuff doesn’t stay in order and it gets lost or it gets duplicated, why on earth would anyone want to build a system like this, first of all, and how could you possibly use it to do anything useful? And the answer is, of course, that you have to do something to put on top of the basic postcard service, something that makes it sequenced and reliable and flow controls. And that’s what the transmission protocol does. It transforms an unreliable service into a reliable one by adding procedures in another layer on top of the basic Internet postcards. Now this notion of layering is pretty important because it is sort of the keystone of the architecture of the Internet. It’s built up in layers, each layer relying on the underlying one for some particular function or service and it works its way down, all the way down to the bottom part where you’re passing photons around in optical fibers, for example.

So let’s see how we would describe how the TCP does its job. Let’s suppose…let me pick on my friend Skip Factor sitting in the middle of the audience there…Skip, let’s suppose that you were trying to send me a novel by this postcard service. It’s the only way you could send it, was by postcards through the postal service. So let’s go through and see how you would design a protocol to deliver the novel reliably to me. Well the first thing you would do, you would have to cut up the pages of the novel and paste them on the postcards so they ‘d fit. And the n you’d notice that not every postcard has a page number because you cut the pages up. And you’d remember that the postcards may get out of order and so it would be a good idea if you ordered every postcard sequentially so I could put them back in order. Otherwise I would have trouble getting the novel put together and read. Then you’d think a little more and you’d say you know, some of these postcards are going to get lost so I better keep copies of them in case I have to re-send them to you to recover from the lost postcard. And then you’d wonder, how do I know when I should send these copies? And then you’d have a bright idea and you’d say, ‘well why don’t I have Vint send me a postcard every once in a while saying "I got everything up to postcard number 206." And that way you’d know you could throw away all the copies up to that point because I had gotten all of those. And then you’d think, "You know, that postcard that Vint’s supposed to send me might get lost, just like all the other postcards might." So that’s not a perfect scheme. And in the end, in addition, you’d decide that after you hadn’t heard anything from me for a while, you’d start sending copies of postcards that had not been acknowledged until you finally hear from me about the ones I have seen. Well that’s basically how the TCP works, except for one other little detail. Let’s suppose that you understand that the mailbox has a finite size. Well mine does anyway. Let’s suppose that you started with a thousand-page novel and you needed to turn that into two thousand postcards. And you troop down to the post office and you gave them all two thousand postcards and by some incredible miracle the post office actually delivered them all on the same day at the same time and they didn’t fit in my mailbox and some of them would fall on the floor and the dog would eat some and the wind would blow some away and you would have to re-transmit them. So might decide to have an agreement with me not to send more than say 100 at a time, until you had gotten some indication of acknowledgement so that you didn’t overload my mailbox. That’s basically all there is to TCP and IP, and that’s basically how the Internet works. Now I’ve left out little details like - ‘how does a router system work’, routing protocols, the domain name system and translation of domain names into IP addresses, but that’s, you know, all little minor detail. Basically that’s all there is to the Internet. It’s a simple system. It’s just that it’s very big and there’s lots of it.

In fact, we’re still in the middle of a kind of "Goldrush" period. And my favorite observation about goldrushes is an important lesson…most people who make money in a goldrush are not the people who are looking for the gold. The people that make money in the goldrush are the people who are selling picks and shovels to the other people who are looking for the gold. And that’s how the telecommunications industry is making money on the Internet. I mean, that’s what we do…we sell the electronic equivalent of picks and shovels to other people who are looking for gold on the Internet and that’s turned out to be quite a good business for us. So we have no complaints about that, but it is important to know that in a goldrush period there is no guarantee of business models that will deliver value. In fact, we are still in the middle, as a community of Internet suppliers and users, of trying to figure out good business models that will in fact produce profits on the net. So there are literally thousands of experiments going on every day as people try out different kinds of business models to see what works and what doesn’t. The good thing is that you can observe what is going on, on the net and you can see what works and what doesn’t. So you don’t have to use up your nickel to make all the mistakes. You can let other people make some mistakes and then you can avoid those yourself. But still we are in the middle of trying to work out as a community what good business models there are.

Now to add to what Bob was saying at the beginning about the growth of the net, this [showing PowerPoint slide] is just a three year snapshot from the middle of ’97 to the middle of the year 2000. About 1.3 million dotcom domain names had been registered, specifically in that top level of domain in July of ’97. 15 million had been registered by July of the year 2000. There were 22.5 million servers on the net in the middle of ’97, that is to say machines with permanent IP addresses, not counting laptops that were getting dynamic addresses assigned as they dialed up through the telephone net, got on for a while, and disconnected. These are permanent server machines. Now there were about 100 million in the middle of this year. Internet services available now in 218 countries, although I’d be the first to remind you that the level of penetration is not uniform. There are some countries with very little service and some with a great deal. China for example has something like 10 or 12 million users, but there are over a billion people in China, so the penetration level is quite small. Whereas in Sweden, for example, something like 70% population has access to the Internet. And incidentally also to cell phones… the Nordic countries tend to be heavily penetrated by Internet and cell phones and no one quite knows why except to speculate that those long, cold Nordic nights and they don’t have anything to do but surf the net and talk on the phone. Now the number of users on the net has grown by over a factor of six in that three year period to 333 million. But to put this in perspective, the number of terminations in the telephone system is almost a billion now. That is about 700 million are wire line terminations, and about 300 million are mobile cell phone. Sometimes it feels like they are all right here in Charlottesville, I know. And that of course is growing very quickly. The cell phone business is growing as much as 50% per year. The telephone or wire line business grows at maybe 5to 10 % per year. The Internet’ s been growing between 80 and 100% per year since 1988, when we started keeping statistics on this, and it has not diminished in that time.

To give you a sense of where people are who use the net, this is an interesting chart because, oh maybe 5 years ago, the Canadian and US figure would have represented 90% of all the users, they were all located in North America. But now, in the year 2000, of the 330 million users, less than half are in North America. So the growth is now picking up outside in the rest of the world. Europe has doubled in the last year, from about 47 million to 91 million users. In Asia, in the Pacific Rim, it’s doubled from 37 to 75 million. It’s a little disappointing though when you figure out the total population in Asia in the Pacific Rim, if you include China and India and Indonesia and Malaysia and Japan, it’s over 2 billion people. And if only 75 million out of 2 billion have Internet, that’s a tiny per capita penetration. Of course it varies…Japan for example has just had a phenomenal increase in Internet use. It’s gone from about 7 million users to 16 million users in the last year, mostly as a consequence of Internet enabled mobile telephones. It’s called bilcamo I-mode, and it’s extremely popular especially among teenagers. Latin America has tripled in the past year from about 5 million users to 13 million. And Africa, this is really a difficult continent to work with, it’s population of Internet users has doubled also in the last year. However most of those users are either in South Africa or Egypt or Tunisia or Morocco and in between there’s very little. Although about 3 weeks ago, the last country that didn’t have Internet access got up on the net. So at least there is some Internet access everywhere in Africa, but there’s not a whole lot. And finally, the Middle East has doubled in the last year. Israel has the bulk of the users, but the surrounding Arab states are beginning to take up the Internet as well, despite the fact that there was a lot of resistance for a while to the regimes in the area that were concerned about exposing their population to content they thought was inappropriate. I think they concluded that: A) you can’t stop the Internet, and B) if you’re not part of the Internet Revolution you may fall behind in the new e-economy. And they concluded that when the oil runs out, they want to be in a position to be part of that new economy. Now just to give you an idea of the lengths that people will go to get hooked up, a few months ago I got an email with a digital picture attached to it from two guys in Kihihi, Uganda. It’s a little city that has no power and no telecommunications, sot hey got a solar panel and a satellite transceiver and a PC. And they hooked it all up and they got on the net and then they sent me an email with this picture attached to show that they had gotten online. Now I am certainly not going to tell you that this means Africa is up on the net, but what it does tell you is that where there is a will there is a way. And there is an increasing amount of interest in using solar power, for example, to do a variety of things, including getting people up on the net. I’ve been watching the user trends now for some time, and my current projections are that by 2010 some 3 billion people, half the world’s population, will actually be online on the Internet.

Now for a while I was trying to track how many servers there would be on the net and the yellow bar chart represents my projections out to the year 2006, when I estimated that there would be about 900 million servers on the net, which would make the net about the same size as today’s telephone system. What I had not understood and didn’t until I went to visit Allen Erickson in Sweden a few months ago, is how many Internet-enabled cell phones there were going to be. And they projected as many as one and a half billion cell phones that would be Internet-enabled, now they said as early as 2004, and I thought I would be a conservative engineer and guess that some of that was hyperbole and marketing hype and so I pushed it out a couple of years. But even so, if you say that there is a billion and a half cell phones online by 2006 plus the 900 million servers, you are at 2 and a half billion roughly devices on the Internet. And anybody who is paying attention to IP version 4 and the address space that’s available and how poorly it’s allocated, knows that 2.4 billion devices would represent a 60% allocation of the 4.2 billion addresses available on IP version 4. So it should be no surprise to you that I am concerned about moving to a larger address space. I’ll come back to that.

Something else that’s important to those of us who run pieces of internet is that we’re being asked to support media that the internet didn’t originally get designed to support, things like video and radio and telephony. Now I guess I should tell you though that it is not new. I mean we’ve been putting these media onto the Internet for a long time, more than 20 years, in fact almost 25 years. I can remember doing an Internet voice experiment on the Internet of the 1975 vintage. Now the backbone of the Internet in 1975 was 50 kilobits per second. That’s the backbone of the network. You get that on a dial-up line today, if you’re lucky. So when we were doing voice communication on the Internet with that kind of backbone, clearly we had to compress the voice signal by a lot. And so we used a scheme called LPC 10, which stood for Linear Predictive Code with 10 parameters. The system modeled the voice track as if it was a stack of cylinders whose diameter varied as a function of time and was excited by a formant frequency. And what you said to the other side was the model of the voice track, with the diameter of these 10 cylinders and the formant frequency that it was being excited by. Well that reduced the data rate to 1800 bits per second, which is pretty good compression down from what would normally be 64 kilobits per second. Then it turns out that there is a side effect to this kind of compression…you lose a lot of information. And it turns out that anyone who speaks through this system comes out sounding like a drunken Norwegian. So the day came when I had to demonstrate the system for a bunch of generals over at the Pentagon. And I’m sure that Dr. Jones will appreciate this story since she served there for some years. We wanted to show them how this thing worked and I remember trying to figure out how I was going to do this and then I remembered that one of the participants in the Internet voice program worked for the Norwegian Defense Research Establishment. So we got Ingvar Lum to be the speaker and it sounded just like him and they never knew the difference. We didn’t tell them that everyone would sound that way.

Well radio, on the other hand, has become very popular. I’m sure many of you have enjoyed listening to sound on the net, to say nothing of downloading. The radio is actually not a big challenge on the net. It’s about 20 to 30 kilobits per second and that’s easily carried on today’s network. There is a little issue about having to buffer the sound, the packets that are showing up that will be interpreted as sound. Because if you don’t do that you may run out of packets before the next one shows up and the sound will break up. So you fill a bucket full of these packets before you start playing them out. What it amounts to is a kind of a sticky radio where you tune into some station and nothing happens for a while and then the sound comes out.

In terms of video, it’s about the same thing. It’s a one-way transmission. You use buffering in order to deal with variable delay and inter-arrival times of the packets. The only thing is that for good quality full screen 30 frame per second video, you need about 400 kilobits per second minimum under good compression practices. And so it’s a good bet that you won’t be watching good quality video on a dial-up line. You need at least 8 tines what a dial-up line can deliver. There are, in fact, technologies that I know all of you know about now – digital subscriber loops and cable modems and MMDS and other alternatives that deliver well in excess of a million bits per second to subscribers on those channels. So it is possible to deliver good quality video and I think you’ll see more and more of that, although that still hasn’t stopped people from trying to do video on a dial-up line. Of course what you get is about a 2-inch screen and maybe you’re lucky if you get 6 or 7 frames per second and a little caption at the bottom that says "Congratulations, you turned your $5000 laptop into a 1928 television set."

Now in addition to carrying these media on the Internet, there is something else that’s going on too. The older media are beginning to carry Internet packets. So digital broadcast satellite, using M-peg and coding can carry Internet packets inside the M-peg frame. So there is nothing to stop us from actually doing multicasting of IP packets through digital broadcast satellite. And in fact some people are beginning to do that. You could carry the stuff on cable, you could carry it on radio, and you could even carry it on broadcast television.

I did want to mention one other interesting thing. There are other modalities that we are starting to experiment with on the net. Handwriting is one and voice recognition is another. The handwriting is interesting because those of you who might remember the Apple Newton that did such a poor job of handwriting recognition that cartoons were written about it describing all the funny interpretations that the software made of your scribbles. I’ve now seen laptops, Palm Pilots and things like that, that are so good at recognizing handwriting that at least several of my friends don’t bother with pads and paper anymore. Now I have this fantasy about putting up machines on the Internet that understand speech. And the reason this gets exciting to me is that…first if I can deliver sound to that machine, by any means at all, whether it’s the telephone system or the Internet or anything else, then that machine can now take input speech, understand it, and then take actions on it. Now if I also have lots and lots of devices on the Internet that are willing to accept control through the network, I can turn any device on the Internet into something which is speech-controllable by putting software into the machine that knows how to understand speech and letting it then take action to send commands through the network to the target. I’ll give you a few examples of that later on. But the idea here is that you can speech-enable almost anything as long as you put it up on the Net and put all the devices on that have to be controlled. And so I find that to be a very intriguing proposition because what that means is that once you put the speech-understanding system and the device under control on the Net, now you can build all kinds of new products and services based on that. So third parties get an opportunity to build on top of that platform and invent new useful things.

Now lest you think that I’ve lost my mind or even if you think that I’ve lost my mind, I should tell you that one reason that I think lots of devices will be on the Net is that people are starting to build the TCPIP protocols into hardware. Last summer two graduate students at the University of Massachusetts built a two-chip web server. One chip did the TCPIP protocols and the other chip did the host or file transfer protocols or the HTTP to deliver traffic to a web browser. The whole she-bang is not much bigger than a couple of quarters. I suspect it costs much more than $0.50, but in fact it’s not very big and you know that over time, when you put things in hardware that eventually the costs come down. The current estimates are that you could do this for less than $5. And so if you could Internet-enable devices for $5 or less, you can imagine that lots of things will happen. And of course I place my faith in the students of the world because once you get a $5 chip set that lets you Internet-enable anything, you can be sure that every student in the world is going to try to Internet-enable almost anything you can think of. Some of it will be silly, but some of these things will actually come out to be quite useful and we’ll be surprised about that. In fact, there are some people who are already trying to make money out of it like the Electrolux folks in Sweden who have Internet-enabled a refrigerator. Now I don’t know about you, but at home I have a family communication system. It’s called a refrigerator and it’s made out of magnets and paper and the thing is just covered with all this stuff…So this would just simply augment the family communications system by adding email and web surfing and other kinds of things. Somebody suggested that we could put a bar code scanner on the refrigerator so that it could see what was being put into the refrigerator and keep track of that. You know it probably wouldn’t know how much milk was left in the jar or the bottle, but it might know when it went in there. So you might get an email or a page from your refrigerator saying that if you don’t do something about it the milk is going to crawl out on its own. But another suggestion was that it could know what’s in there and so it could be searching around the Internet for you to tell you what recipes you could make with whatever is in the refrigerator. Or it might page you while you’re at the store and say, "Don’t forget to pick up the marinara sauce. I have everything you need for spaghetti except for that." So you might get a little help from the refrigerator preparing interesting meals.

Now in Japan they showed me an Internet-enabled bathroom scale. You know you step on the scale in the morning and it sends your weight to the doctor and that becomes part of your medical record. It seems like a reasonable thing to do. But then somebody asked me, "What happens if the refrigerator gets the same information? You know, and then it refuses to open because you’re on a diet."

Well I couldn’t do this without mentioning just a little bit about the economic impact of the Internet. These are some of the estimates for the total amount of e-commerce on the Internet in just four years time. The current estimates now are almost $6.8 trillion. Now keep in mind that that’s about 20% of the gross world product four years from now. So that means that the Internet may be carrying one fifth of all the world’s commerce in just four years time. Now that should scare you. Some of you should be levitating out of seats because that means that the Internet had better be even more reliable than it is now. Maybe a lot more reliable than it is now. It means the software that implements all this stuff had better be a lot more reliable too. And it means that we had better learn how to secure and make private some of the transaction content. So you have a lot of work ahead of you, those of you who are interested in getting involved in the Internet because this isn’t going to wait. We have to do something about all those requirements.