Tracking the Internet into the 21st Century with Vint Cerf

Uploaded by Google on 07.12.2007


We'll get things kicked off.
So first of all, welcome.
And to all the Americans in the house, having

A few.
All right.
All right.
So I'm Randy Knaflic.
I head up Google's recruiting machine here in EMEA, and I'm
based here in Zurich.
And I just wanted to--
I get the pleasure and the honor of introducing our guest
this evening.
And since we have all of you here, I thought it would be
good chance to just spend just a couple minutes because some
of you may or may not be familiar with this search
company called Google.
So I'll do my best to give you a little education on that.
But probably it's better just to tell you little bit about
Google here in Zurich, Switzerland.
So we've actually been here for nearly four years now.
And started with actually one engineer.
And we have the host of having, actually, that one
engineer in the front row.
So if you're interested in seeing how you go from one
engineer to a site that's over 350 in a very short time, he
can tell you how it's expanded quickly.
And I can tell you--
I had a little bit more hair when this all started.
So Google's idea was to basically take the model that
worked really well in Mountain View, the engineering model
that worked there, and basically recognizing that
there's this one little fact.
That is, there are a lot of engineers in the world.
And guess what?
They don't all sit in Mountain View, California.
So, believe it or not, they decided they wanted to bring
this model a little bit closer to the masses.
And Zurich was the first site where they decided to do that.
And since then, we've actually grown to having 12 engineering
sites throughout EMEA and a number of different locations
in all the major cities that have engineering talent.
And it's worked really well.
And so having the engineering sites in these locations is
enabling us to do things both for the global market and for
the local market.
So most of you are here in Zurich.
And whether you knew this or not, next time you're on
Google, check out Google Maps and all those really cool
things that you can do like see what time the bus--
the bus outside the
Hurlimann-Areal that comes every--
I think it's, what, four days?
It's a really bad schedule on that.
But you would see that schedule.
And that was done by our engineers here.
Not not the bus schedule, the letting you know
when it will come.
We're working on that as well.
We'll get back to you on that.
So we're continuing to grow.
And we'll continue to tap into the talent that is in not only
Switzerland, but beyond.
This site is very special, and it's probably no secret that
Zurich is a great place to live.
And we're able to track a lot of talent from all over.
In fact, we have over 42 different countries
represented just in this one location.
Actually, I should say two locations, because this is now
our second site that all of you will have an opportunity--
those who opted for a tour will get to see our new site
here in the Hurlimann-Areal.
So that leads us to the next step, which is actually we
have a returning visitor and our guest and honorable guest
for this evening.
And it's fun, because every time Vint comes back, he gets
to see how-- he comes.
There's 30, 40 engineers here.
He runs away.
Comes back, there's 300 and some here.
So we're going to keep doing that.
We're hoping now the next time he comes back, we even have
this entire building filled and we'll continue to grow.
So he's a man who needs very little introduction.
He's actually the real father of the internet, despite what
some Americans think about Mr. Gore.
The work that he did in creating the TCP/IP protocols
is obviously what has enabled companies like Google to
exist. So when he joined in September of 2005, this was a
huge honor for us and a great win to have such a
visionary on board.
And we're very pleased to have him.
And it's with my great pleasure that I
introduce Vint Cerf.

VINT CERF: Thank you very, very much.
And I really appreciate everyone's taking time out,
especially in the middle of the week, in the
evening, to join us.
It's a real pleasure to have you here on our Google campus.
I don't know where Renee LaPlante is right now, but
it's her birthday today.
Renee, where are you?
Happy birthday.

As as for the crack about Al Gore, there's always some
nincompoop who brings that up.
Al Gore deserves credit for what he did as a senator and
as vice-president.
He actually helped to pass legislation that enabled
NSFNET backbone to grow and to permit commercial traffic to
flow on the government-sponsored
backbones in the US.
Had he not done that, it's pretty likely that the
commercial sector would not have seen an opportunity to
create a commercial internet that all of us can enjoy.
So he does deserve some credit for what he's done.
I meant to start out--
And that's all the German you're going to get tonight.
I did spend a very pleasant six months in the area around
Stuttgart as a young student at Stanford University.
Well, my purpose tonight is to give you a sense for what's
happening to the internet today and where we think it's
headed in the future.
And I thought I would also take advantage of the time to
give you little glimpses of what it was like in the early
stages of the internet.
But first, let me explain something about my title.
I'm Google's Chief Internet Evangelist. It wasn't a title
that I asked for.
When I joined the company, they said,
what title you want?
And I suggested Archduke.
And they said, well, that doesn't quite fit in any of
our nomenclature.
And they pointed out that the previous Archduke was
Ferdinand, and he was assassinated in 1914, and it
started World War I. So maybe that's not a title that you
want to have. And they suggested that, considering
what I've been doing for the last 35 years, that I ought to
become the Internet Evangelist for Google.
So I showed up on first day of work wearing this outfit that
I guess you see over on your right.
This is the formal academic robes of the University of the
Balearic Islands.
And it was the most ecclesiastical
outfit that I owned.
So I showed up wearing that on my first
day of work at Google.
And Eric Schmidt took this picture.
It's not often you can find an opportunity to wear something
looking like that.
So I took advantage of that for that one day.
Well, let me just start out by reminding you of some
statistics of the internet over the last 10 years.
10 years ago, I would have been very excited to tell you
there were 22 million machines on the internet.
Now, there are almost 500 million million.
And these are servers, the web servers, the email servers,
and the like.
It's not the machines that are episodically connected, like
laptops or personal digital assistants.
The number of users on the net has grown to 1.2 billion,
which sounds like a big number until you realize that there
are 6 and 1/2 billion people in the world.
So the chief internet evangelist has 5.3 billion
people to convert to internet use.
So I have a long ways to go.
The other thing which has been happening in the telecom
environment over the last decade has been the rapid
influx of mobiles.
The estimates now are that there are 3 billion by the end
of this year that are in use, 2.3 billion accounts, and 3
billion mobiles, which means there are 700 million people
with more than one mobile.
What's important to us and others in the internet
environment is that many people will have their first
introduction to the internet through a mobile and not
through a laptop or a desktop.
There are estimated to be about 10% of all the mobiles
in use that are internet enabled.
And so as time goes on and more and more of these devices
become part of the landscape, an increasing number of people
in the world will have their first introduction to the
internet by way of a mobile as opposed to other instruments.
If we look at the distribution of users on the network, the
first thing that strikes me, anyway, is that 10 years ago,
North America would have been the largest absolute
population of internet users.
But today, it's Asia, which includes China, and India, and
Indonesia, and Malaysia, Japan, and so on.
But interestingly enough, this large number, 460 million
people, represent only 12% of the population in that region.
So as they reach the same penetrations as we have, for
example, in Europe, at 42%, the absolute numbers of Asian
users will increase.
This tells you something about what to expect in terms of the
content of the internet.
The kinds of interests that people will have, the culture
and style of use of the net will all be strongly
influenced by our colleagues living in that region.
Europe has almost 338 million users, with a
penetration of about 42%.
I've given up trying to make any predictions at all about
Europe because you keep re-defining yourselves by
adding countries to.
So whatever happens is going to happen.
Africa is at the bottom of the list here in terms of the
percentage penetration.
There are a billion people living in Africa, but very few
of them, 40 million of them, have access to the internet.
It's a big challenge there.
The telecom infrastructures are still fairly immature.
The economies very pretty dramatically.
And so getting them up and running on the internet is an
important task, and one which is a significant challenge.
I thought you'd find it amusing to go way back into
history see the beginnings of the predecessor to the
internet, the ARPANET.
There was a four node system that was set up
in December of 1969.
And I was fortunate enough to be a graduate student at UCLA
in September of 1969.
I programmed the software that connected a sigma seven
computer up to the first node of the ARPANET.
The sigma seven is now in a museum.
And some people think I should be there too.
But this was the beginning of wide area packet switching.
And it was a grand experiment to see whether or not that
technology would actually support rapid fire
interactions among time shared machines.
And indeed, it worked out.
The packet switch was called an IMP, an
Interface Message Processor.
This is what it looked like.
It was delivered by Bolt, Baranek, and Newman, the
company in Cambridge, Massachusetts, in a very, very
heavy-duty metal box.
They knew that this was a military contract.
And they didn't know whether they would be dropping these
things out of helicopters or what.
So they put it in a very, very heavy duty container.
Considering it was installed at UCLA surrounded by graduate
and undergraduate students, probably this heavy-duty
container was exactly the right thing, even if it never
deployed into any other place.
This picture was actually taken in 1994.
It was the 25th anniversary of the ARPANET.
The guy on the far left is Jon Postel, who all by himself
managed the RFC series as the editor, managed the allocation
of IP address space, and managed the delegation of top
level domains in the domain name system for over 25 years.
You can imagine, though, by the time 1996 rolled around,
the beginnings of the dot boom have happened, Netscape
Communications has done its IPO, the general public has
discovered internet, and we're off and running, Jon realised
that that function that he performed needed to be
And so he began a process of trying to figure out how to
create an organization that would perform these functions.
In the end, an organization called ICANN, the Internet
Corporation for Assigned Names and Numbers, was created in
1998 to take over the responsibilities of domain
name management, internet address allocation, and the
maintenance of all the protocol tables associated
with the internet protocols.
Sadly, Jon Postel passed away two weeks before ICANN was
actually formally created.
But it has now persisted since 1998.
I had the honor of serving as chairman of the board of ICANN
until just about two weeks ago.
And I managed to escape because my sentence was up.
There are term limits in the bylaws.
And it said I couldn't serve anymore on the board.
And, frankly, I was pleased to turn this over to my
successor, a man named Peter Dengate Thrush, who is from
New Zealand.
So the point I want to make here, apart from the amusing
diagram of 1994, is that the internet it is now and has
always had a certain international character to it.
The guy in the middle is Steve Crocker.
He was the man who really blazed trails in computer
network protocols.
He ran what was called the Network Working Group.
which was a collection of graduate students in various
in computer science departments around the United
States, developing the first host to host protocols.
It was called NCP, Network Communications Program.
And Steve was the primary leader that
developed that protocol.
It was used until 1983, when the TCP/IP protocols were
introduced to a multi-network system.
You'll notice that we tried to demonstrate in this picture
for Newsweek how primitive computer
communications was in the 1960s.
It took us almost eight hours to set up this shot.
And we drew all these pictures on foolscap paper of the
clouds of networks.
And then we had to buy some zucchinis and yellow squash
and string them together.
You'll notice that this network never would work
because it was either ear to ear or mouth to mouth.
But there was no mouth to ear.
We posed it this way on purpose, hoping there would be
a few geek readers of Newsweek who would get the joke.
This is what I looked like when the TCP/IP protocols were
being developed.
I demonstrated--
not TCP/IP but the ARPANET from South Africa in 1974.
This was actually an interesting experience because
we brought an acoustic coupler with us into South Africa.
It was the first time that the South African telecom company
had ever allowed a foreign object to be connected to
their telephone system.
And they were very concerned that this acoustic coupler
might do damage to their network.
So we managed to persuade them that it would be OK.
And we connected the terminals in South Africa to the ARPANET
by way of a satellite link all the way back to New York at
the blazing speed of 300 bits per second.
Well, the internet got started in part because Bob Kahn, the
other half of the basic design of the internet, told me that
in the Defense Department, he was looking at how to do
computers in command and control.
And if you were really serious about putting computers where
the military needed to be, you had to have computers running
in mechanized infantry vehicles, and in tanks, and
other things.
And you couldn't pull wires behind them, because the tanks
would run over the wires and break them.
So you needed radio for that.
And you also needed to have ships at sea communicating
with each other.
And since they couldn't pull cables behind them, because
they'd get tangled in knots, instead you needed to have
satellite communication for wide area linkages.
And, of course, we needed wireline communications for
fixed installations.
So there were three networks that were part of the internet
One was called the packet radio net for
mobile ground radio.
Another was packet satellite using Intelsat 4A across the
Atlantic to permit multiple ground stations to compete for
access to a shared communications channel.
And then the ARPANET, the predecessor, which was based
on wirelines.
SRI International ran the packet radio test bed in the
San Francisco Bay area during the 1970s.
And they built this nondescript panel van as a way
of testing packet radio by driving up and down the
Bayshore Freeway and occasionally stopping to make
detailed measurements of packet loss, signal to noise
ratio, the effects of shot noise from cars running back
and forth nearby the van.
The story goes that one day they'd pulled off to the side
of the road, and the driver, who was another engineer, got
out from the cab, and went around, and got into the back
of the van.
They were making a bunch of measurements.
And some police car pulled up and noticed that there was
nobody in the cab.
So he went around and knocked on the door.
And, of course, they opened up the door.
And this policeman looks in.
And he sees a bunch of hairy, geeky people with computers,
and displays, and radios, and everything else.
And he says, who are you?
And somebody says, oh, we work for the government.
And he looks at him, and he says, which government?

But officer, we were only going 50 kilobits per second.
Well, I remember in 1979, after we demonstrated that
this technology actually worked, that we wanted to
convince the US Army that they ought to seriously try these
out in field exercises.
So I had a bunch of guys from Fort Bragg 18th Airborne Corps
coming out to drive around on the Bayshore to actually see
how it worked.
And later, we actually deployed packet radios at Fort
Bragg for field testing.
They were big one cubic foot devices that
cost $50,000 each.
They ran at 100 kilobits and 400 kilobits a second.
They used spread-spectrum communications.
Now, this is pretty advanced considering it's 1975.
And, of course, the physical size of the radio tells you
something about the nature of the electronics that were
available at the time to implement this.
This is a view of the inside of that van.
Something else that was very interesting in all of this is
that some of you will be familiar with voice over IP.
Maybe many of you are using Google Talk, or Skype, or one
of the other applications, iChat.
We were testing packetized speech in the mid-1970s.
So, in fact, this is not such a new thing, after all.
In the case of the packet radio network and the ARPANET,
we were trying to put a packetized speech over what
turned out to be a 50 kilobit backbone in the ARPANET.
And all of you know that when you digitize speech, normally
it's a 64 kilobit stream.
So cramming 64 kilobits per second into a 50 kilobit
channel is a little bit difficult.
And in fact, we wanted to carry more than one voice
stream in this backbone.
So we compressed the voice down to 1,800 bits per second
using what was called linear predictive code with 10
All that means is it the voice track was modeled as a stack
of 10 cylinders whose diameter was changing as the voice
would speak.
And this stack was excited by a Formant frequency.
You would send only the diameters of each of the
cylinders plus the Formant frequency to the other side.
You'd do an inverse calculation to produce sound
and hope that somehow it would be intelligible
on the other end.
Speaking of intelligible, I am struggling right now with a--

it's vodka.
This should be very interesting by the end of it.
So part of the problem with this compression ratio, going
from 64 kilobits down to 1,800 bits per second, is that you
lose a certain amount of quality of voice.
And so when you spoke through this system, it basically made
everyone sound like a drunken Norwegian.

The day came when I had to-- sorry about this.

The day came when I had to demonstrate this to a bunch of
generals in the Pentagon.
And I got to thinking, how am I going to do this?
And then I remembered that one of the guys that was
participating in this experiment was from the
Norwegian Defense Research Establishment.
His name was his name was Yngvar Lundh.
And so we got the idea that we'd have him speak through
the ordinary telephone system.
Then we'd have him speak through the
packet radio system.
And it sounded exactly the same.
So we didn't tell the generals that everyone would sound that
way if they went through the system.
Today is actually a very important milestone.
Today is the 30th anniversary of the first demonstration of
getting all three of the original networks of the
internet to interconnect and communicate with each other.
We took-- the packet radio van was driving up and down the
Bayshore Freeway radiating packets.
They were intended to be delivered to USC Information
Sciences Institute in Marina Del Rey, California, which is
just to the west of Los Angeles.
But we jiggered the gateways so that the routing would
actually go from the packet radio net, through the
ARPANET, through an internal satellite hop down to Kjeller,
Norway, then down by landline to University College London,
then out of the ARPANET through another gateway, up
through a satellite ground station at Goonhilly Downs.
And then up through the Intelsat 4 satellite, then
down to [? Ekham ?], West Virginia to another satellite
ground station, through another gateway, and back into
the ARPANET, and then all the way down to USC ISI.
So as the crow flies, the packets were only going 400
hundred miles, from San Francisco down to Los Angeles.
But if you actually measured where the packet when, it went
over 88,000 miles.
Because it went through two satellite hops up and down,
and then across the Atlantic Ocean twice, and across the
United States.
So we were all pretty excited about the fact that it
actually worked.
I don't know about you, but I've been in the software game
for a very long time.
Software never works.
I mean, it's just a miracle whatever it works.
So we were leaping up and down, screaming, it works, it
works, as if it couldn't possibly have worked.
So we celebrated this particular anniversary a
couple of weeks ago at SRI International.
We invited everybody who had been involved in this
particular demonstration.
And quite a few people were able to come back.
And we got to renew our old acquaintances.
But that was a very important milestone today, 30 years ago.
Of course, if you look at the internet in 1999, ninety even
today, this is the sort of thing you see.
Highly connected, much larger, more colorful.
And that's about as much as you can say about the internet
that it's a pretty accurate description.
It got a heck of a lot bigger over the 30-year period.
Some of the things that have made the internet successful
were fundamental decisions that Bob Kahn and I and others
made at the beginning of this thing.
One thing that we knew is that we didn't know what new
switching and transmission technologies would be invented
after we had settled on the design of the internet.
And we did not want the net to be outmoded.
We wanted to be future-proof.
So we said we don't want the internet layer protocol to be
very aware of or dependent upon which technology was used
to move packets from one point to another.
We were fond of observing that all we needed from the
underlying transmission system is the ability to deliver a
bag of bits from point A to point B with some probability
greater than zero.
That's all we asked.
Everything else was done on an end to end basis using things
TCP or UDP in order to recover from failures, or to
retransmit, or to weed out duplicates.
So we were, I think, very well-served by making that
particular philosophical decision.
But there was something else that also derived from that
decision that I didn't fully appreciate until later.
The packets not only don't care how they are being
carried, but they don't know what they're carrying.
They're basically ignorant of anything except that they're
carrying a bag of bits.
The interpretation of what's in the packets occurs at the
edges of the net, in the computers that are
transmitting and receiving the data.
The consequence of this end to end principle has been that
people can introduce new applications in the internet
environment without having to change the underlying networks
and without having to get permission from the internet
service providers to try out their new ideas.
So when Larry and Sergey started Google, they didn't
have to get permission from an ISP in order to
try this idea out.
Or when Jeff Bezos did Amazon, or when David Filo and Jerry
Yang set up Yahoo!
They simply did it.
And the same is true more recently
when Skype was created.
Nobody had to give any permission to anyone.
Same is true for BitTorrent and many of the other
peer-to-peer applications.
Essentially, you get to do what you want to do because
the underlying network, is at least up until now, very
neutral about the applications.
So this end to end principle has been an important engine
for innovation in the internet in the past. Google and others
believe that it should continue to be an engine of
And it can only do that if the internet service providers
essentially keep their hands off the applications and just
simply carry bits from point A to point B. This doesn't mean
that an ISP can't also have value added applications.
That's not what this means.
It just means that the provider of the underlying
transmission, in many cases broadband transmission, that
provider should not take advantage of carrying the
underlying transmission to interfere with other parties
competing at higher layers in the protocol for applications
that might be of interest to the consumers.
Similarly, the consumers, who believe that they're buying
access to the full internet everywhere in the world when
they acquire broadband access to the net, have a reason to
expect that no matter where they aim their packets, that
the underlying system will carry them there in a
nondiscriminatory way.
This does not mean, for example, that you must treat
every single packet on the internet precisely in
identically in the same way.
We all understand the need for control
traffic with high priority.
We understand the possibility that some
traffic needs low latency.
We understand that you may want to charge more for higher
capacity at the edges of the net.
Net neutrality does not mean that everything is
precisely the same.
But what it does mean is that there is no discrimination
with regard to whose services the consumer is trying to get
to or who is offering those services when it comes to
traversing the broadband channels that
the ISPs are providing.
One other thing about broadband that's turning out
to be a problem is that, at least in the United States,
it's an asymmetric service, as you often can download things
faster than you can upload them, leading to anomalies
like you can receive high-quality video, but you
can't generate it.
My general belief is that the consumers are going to be
unsatisfied with this asymmetry and that there will
be pressure to provide for uniform and symmetric
broadband capacity, which is what you can get in other
parts of the world.
In Kyoto, you can get a gigabit per second access to
the internet.
It's full duplex, and it costs 8,700 yen a month.
It almost made me want to move to Kyoto, because it just
seemed like such a very friendly environment to try
new things out.
So we're very concerned about the symmetry and neutrality of
the underlying network.
The other thing I wanted to point out is that this chart,
which was prepared by Geoff Huston, who's an engineer in
Australia, is intended to illustrate the utilization of
the IP version 4 address space.
The important part of this church is the one that's
trending downward.
That part is saying basically these are the address blocks
that the Internet Assigned Numbers Authority is
allocating to the regional internet registries and that
it will run out of IPv4 address blocks somewhere
around the middle of 2010.
The regional internet registries--
yours here in Europe is the RIPE NCC--
will presumably hand out portions of
those address blocks.
They're likely to use those up by the middle of 2011.
The implication of this is there won't be any more
available IPv4 address space.
This doesn't mean that the network will come to a
grinding halt.
But what it does mean is there won't be any more IPv4 address
space and that the only addresses that will be
available for further expansion
will be IPv6 addresses.
Now, I have to admit that I'm personally the
cause of this problem.

Around 1977, there had been a year's worth of debate among
the various engineers working on the internet design about
how big the address space should be for this experiment.
And one group argued for 32 bits, another for 128 bits,
and another for variable length.
Well, the variable length guys got killed off right away
because the programmers said they didn't want variable
length headers because it was hard to find all fields and
you had to add extra cycles to do it.
And it's hard enough to get throughput
anyway, so don't do that.
The 128 bit guys were saying, we're going to need a lot of
address space.
And the others guys were saying, wait a minute.
This is an experiment.
32 bits gives you 4.3 billion terminations.
And how many terminations do you need to do an experiment?
Even the Defense Department wasn't going to buy 4.3
billion of anything in order to demonstrate this
So they couldn't make up their minds.
And I was the program manager at the time spending money on
getting this thing going.
And finally, I said, OK, you guys can't make up your mind.
It's 32 bits.
That's it.
We're done.
Let's go on.
Well, if I could redo it, of course, I'd go back and say,
let's do 128.
But at the time, it would have been silly.
We were using full duplex Echoplex kinds of interactions
with timeshared machines and local
terminals across the network.
And you can imagine sending one character with 256 bits of
overhead just for the addressing, it
would have been silly.
So we ended up with a 32 bit address space.
I thought we would demonstrate the capability of the internet
and that we would be convinced, if it worked, that
we should then re-engineer for production.
Well, we never got to re-engineer it.
It just kept growing.
So here we are.
We're running out.
And we have to use the new IPv6.
By the way, if you're counting, and you wonder, OK,
IPv4, IPv6, what happened to IPv5?
The answer is it was an experiment in a different
packet format for streaming audio and video.
And it led to a cul de sac and we abandoned it.
And the next available protocol ID was six.
So that's why we have IPv6.
Now, with 128 bits of address space, you can have up to 3.4
times 10 the 38th unique terminations.
I used to go around saying that means that every electron
in the universe can have its own web page if it wants to
until I got an email from somebody at CalTech: Dear Dr.
Cerf, you jerk, there's 10 to the 88th electrons in the
universe, and you're off by 50 orders of magnitude.
So I don't say that anymore.
But it is enough address space to last until after I'm dead,
and then it's somebody else's problem.

I won't have time to go through very many of these,
but I wanted to just emphasize that despite the fact that the
internet has been around for some time now, certainly from
the conceptual point of view for 35 years, there are still
a whole bunch of research problems that
have not been solved.
Let me just pick on a couple of them that I consider to be
major issues.
Security is clearly a huge issue.
We have browsers that are easily Penetrated They
download bad Java code and turn their machines into
zombies which become part of bot nets.
You have problems of denial service attacks.
You have the ability to actually abuse the domain name
system and turn some of its components into amplifiers
denial of service attacks, which adds insult to injury.
Multihoming, we haven't done a very good job of that in
either v4 or v6.
Multipath routing, we usually pick the best path we can, but
it's only one path.
If there were multiple paths between a source and
destination, if we could run traffic on both of them, we'd
get higher capacity.
We don't do that.
We don't use broadcast media at all well in the current
internet architecture.
When you think about it, we turn broadcast channels into
point to point links.
And it's a terrible waste if your intent is to deliver the
same thing to a large number of receivers.
And that could very well turn out to be a useful capability,
not just for delivering things like video or audio to a large
number of recipients, but software.
People want to download a
particular a piece of software.
If enough people wanted the same thing, you could schedule
a transmission over a broadcast channel that would
allow everyone to receive it efficiently, perhaps by
satellite, or over a coaxial cable, or over a cable
television network.
So we haven't done any of those things very well.
And we don't have a lot of experience with IPv6.
And what's even more important, we don't have a lot
of experience running two IP protocols at the same time in
the same network, which is what we are
going to have to do.
So in order to transition, we can't simply throw a switch
and say, tomorrow we're using IPv6 only.
We're going to have to spend years running both v4 and v6
at the same time.
And we don't have a lot of experience with that.
When you do two things instead of one thing, you get more
possible complications.
The network management systems may not know what to do when
it gets errors from both v4 and v6.
Or worse, it gets errors from v6 but not from v4.
The routing is working for one, but not for the other.
Do I reboot the router or not?
What do I do?
So there are a wide range of issues, including a very
fundamental problem with IPv6.
We don't have a fully connected IPv6 network.
What we have is islands of IPv6.
This is not the circumstance we had with
the original internet.
Every time we added another network, it was v5, and it
connected to an already connected network.
But when we start putting in v6 and not implementing it
uniformly everywhere, then we're going to have islands.
They could be connected by tunnels through v4.
It's a very awkward proposition.
Until we have assurance that we have a fully connected IPv6
network, people are going to be doing domain name lookups,
getting IPv6 addresses, trying to get to them, and not
getting there because you're in a part of the internet the
doesn't connect to the other parts of the internet that are
running IPv6.
So these are just headaches that are going to hit.
And they're going to start hitting in 2008 because of us
are going to have to start getting v6 six into operation
before we actually run out of IPv4 four addresses.
Switching to a slightly different view of the net,
there have been some really surprising social and economic
effects on the net that are becoming more visible.
The one that I find the most dramatic is that the
information consumers are now becoming
the information producers.
So you can see it in the form of blogging or Youtube or
Google Video uploads, personal web pages, and other things.
People are pushing information into the network as well as
pulling it out.
This is unlike any broadcast or mass medium in the past. In
the past, a mass medium had a small number of information
producers and a very large number
of information consumers.
The internet inverts all of that and allows the consumers
also to produce content.
Wikipedia has taught us another very interesting thing
about this internet environment.
I want you to think about a paragraph in Wikipedia that
you're reading.
And you see one word which should be changed because
you're an expert in the area and you know that the
statement is wrong or the sense of the
paragraph is wrong.
You could certainly make that one word change.
You would never publish a one word scholarly paper.
You wouldn't publish a one word book.
But you can publish one word of a change
in Wikipedia paragraph.
And it's useful.
It's a contribution to everyone who
looks at that paragraph.
So the internet will absorb the one word change, the one
page change, one paper, one book, one movie, one video.
It is willing to absorb information at all scales and
in all formats, as long as they can be digitized.
So the barrier to contribution into the internet environment
is essentially zero.
Another phenomenon which is rapidly evolving is social
Many of you may already be using LinkedIn or Myspace or
Facebook or Orkut or some of the others.
That's a phenomenon that's going to continue to grow.
Especially young people enjoy interacting with each other in
this new medium.
And they show a considerable amount of creativity in
inventing new ways of interacting with each other.
Similarly, game playing.
Second Life, World of Warcraft,
and a bunch of others.
EverQuest is another one.
What's interesting about these particular environments is
really twofold.
One of them is that there are real people making decisions
in these games.
And some economists at Harvard, for example, have
asked their students to go become participants in Second
Life in order to observe the kinds of economic decisions
that people are making in the context of these games,
because they're actually trying out different economic
principles within various parts of the game environment.
And so it's actually in an experiment that you couldn't
necessarily conduct in the real world that's being
conducted in this artificial environment.
The other important observation I would make is
that the economics of digital information are dramatically
different from the economics of paper or
other physical media.
Just to emphasize this, let me give you a little story.
I bought two terabytes of disk memory a few months ago for
about $600 for use at home.
And I remembered buying a 10 megabyte disk drive in 1979
for $1,000.
And I got to thinking, what would have happened if I'd
tried to buy a terabyte of memory in 1979?
And when you do the math, it would have
cost me $100 million.
I didn't have $100 million in 1979.
And to be honest with you, I don't have
$100 million now either.
But if I'd had $100 million in 1979, I'm pretty sure my wife
wouldn't let me buy $100 million worth of disk drives.
She would have had a better thing to do with it.
The point I want to make, though, is that that's a very
dramatic drop in the cost of disk storage.
You're seeing similar kinds of drops in the cost of moving
bits and processing bits electronically.
The business models that you can build around those
economics are very different from the business models that
were built around other media, paper or other physical media.
And companies that built their businesses around the older
economics are going to have to learn to adapt to the new
economics of online, real time, digital processing
transmission and storage.
And if they don't figure out how to adapt to that, they'll
be subject to Darwinian principles.
This is a very simple principle, adapt or die.
And so if you don't figure out how to adapt, the other choice
is the only one you have. So a number of companies are going
to be, I would say, challenged to understand that the
economics of digital information are really
demanding them to rethink their business models.

This was a chart that was generated by a
company called Sandvine.
They're doing some deep packet inspection to understand the
behavior of users at the edge of the network on
a particular channel.
What they were looking at here is a variety of applications
that are visible as the packets are traversing back
and forth over access lines.
What was important here is that the Youtube traffic
represented somewhere between 5% and 10% of all the traffic
that they measured on this particular access channel.
And I bring this up primarily to say that Youtube is only
two years old.
So just look at what happened with an application very
recently suddenly blossoming into a fairly highly demanding
application in terms of capacity on the network.
We can easily imagine that other applications will be
invented that may have different profiles of demand
for traffic, either uploading or downloading.
or maybe low latency, or many other things.
So the point here is that the network is very dynamic.
It is constantly changing.
New applications are coming along, making new demands on
its capacity.
And so this is not stable in the same sense that the
telephone network was stable, where you could use Erlang
formulas to predict how many lines you needed to keep
people from getting a busy signal, below 1% probability.
The internet does not have stable statistics like that.
And because new applications can be invented simply by
writing a new piece of software, I think we're not
ever going to be able to predict very well the actual
behavior of the net at the edge.
In the core of the net, it's a different story, because
you're aggregating a large number of flows.
And you can get fairly stable statistics for the core of the
net, but not at the edge.
I've been thinking a lot about how the economics of digital
storage and transmission have an effect on certain kinds of
media, like video in particular.
Let's just take a moment a couple of observations.
15% percent of all the video that people watch
is real time video.
It's being produced in real time.
It's a news program.
It's an emergency or maybe a sporting event.
85% five percent of video that people watch is actually
pre-recorded material.
So in this chart, there are two kinds of video, RT for
Real Time video that's been generated in real time.
And PR for Pre-Recorded video that's being transmitted
through the network.
And imagine now that we've got two axes.
One is the transmission rate that's available
to you as the consumer.
And the other is the storage you have available locally.
And the split is that high transmission rate means it's
sufficiently high to deliver things in real time and low
means you can't deliver it in real time.
Low storage means there isn't enough memory locally to store
any reasonable amount of video.
High means there's enough storage to store reasonable
amounts, which might be measured in hours of video.
So the question is, which quadrant are
you in as the consumer?
If you're in the lower left hand quadrant, where you can't
transmit in real time, and you don't have any place to store
it, you're basically out of luck.
Video is not an interesting medium for you.
If you have very high transmission rates and no
storage available, you can easily receive the streams in
real time, just as you would over a typical cable or
satellite or over the air transmission system.
You can even potentially receive the pre-recorded
material at higher than real time speeds.
But since you don't have any place to store them, it
doesn't do you a lot of good.
So basically, in the upper left, you're stuck with
streaming video in real time.
In the upper right, it's much more interesting.
Because here you have high speed available and you have a
lot of storage available.
The real time stream could be delivered and
watched in real time.
It could be delivered in real time and stored away and
watched later, just like TiVo or other
personal video recorders.
But it's the pre-recorded stuff that gets really
You could clearly transmit it in real time.
But you can also transmit it faster than real time because
you have a data rate that exceeds the rate at which the
video is normally transmitted for viewing.
The implication of this is that video on demand no longer
means streaming video.
It means delivering video potentially faster than you
could watch it.
Anybody that uses an iPod today is experiencing that.
Because you're downloading music faster than you could
listen to it.
And then you play it back at your leisure
whenever you want to.
It's my belief that IPTV is going to be the download and
playback style of iPod as long as the data rates at the edges
of the net are sufficiently high.
So what does that mean for the television industry?
And I'd like to use the word television here to refer to a
business model and the word video to refer to the medium.
And so my interest here is understanding what happens to
the video medium and the business of video when it ends
up into an internet environment.
One thing that's very clear is that because you packetized
everything when you're downloading, the data that
you're downloading doesn't have to be confined
to video and audio.
It could easily contain other information.
So when you get a DVD, it has bonus components on it.
It's bonus videos.
It's textual material.
Maybe it's the biographies of the actors, or the story of
how the movie was made, or the book that the
movie was based on.
So when we're downloading stuff associated with video
and it's coming through the internet, we can download all
forms of digitized content, store it away, and
then access it later.
Among the things that could be downloaded
is advertising material.
In the conventional video world you, interrupt the video
for an an advertisement, and you force it on the users, on
the consumers.
In the world of internet based systems, when you're playing
back the recorded content, there's a program, a computer,
which is interpreting the traffic and
interpreting the data.
And so it's not just a stupid raster scan device.
It can actually make decisions based on the kind of
information that's being pulled up.
So imagine that you've composed an entertainment
video and that you've made some of the objects in the
field of view--
like maybe this Macintosh is sitting in the field of view--
you've made those objects detectable or
sensitive to mousing.
So if you mouse over that particular object, it
And a window pops open.
It says, gee, I see you're looking at the Macintosh.
Let me tell you a little bit more about that product.
Click here if you'd like to find out whether there are any
available at the Apple Store.
By the way, do you want to complete the transaction now?
And then go back to watching the movie.
The idea of allowing the users to mouse around in the field
of view of an entertainment video it is a transforming
idea with regard to advertising.
And it feels a little funny, a computer programmer like me
sitting up here, getting excited about advertising.
But remember, that's where Google makes all its revenue.
So we care a lot about new styles of advertising that
would improve the consumers' control over what advertising
he or she has to be exposed to.
And also, it turns out the advertisers care a lot about
knowing whether the users are interested in their products.
And so knowing that nobody is--
if they're not interested, they don't click.
If they are, they do.
And that is a big jump up in understanding something about
the potential client for your products or your services.
So my prediction is that video in the internet environment,
where high speed interfaces are available and lots of
storage are available, will be a transforming opportunity for
users to control advertising and for advertisers to wind up
with a much better product than they have today.
I mentioned mobiles before.
And I just want to emphasize that these are
programmable devices.
These are not just telephones anymore.
Google recently announced an operating system called
Android that we would like to make available to anyone who's
building these wireless platforms. | purpose is to
open up the platform so that you can download new
applications and allow users to try out new things without
too much difficulty.
These things are already useful for accessing
information on the net.
Here, especially, in Europe, they are being
used to make payments.
This is a challenge, though.
I've got a BlackBerry here, and it has a screen that's the
size of a 1928 television set.
And the data rates that you can reach this thing with vary
from tens of kilobits a second to maybe as much as a megabit.
And the keyboard is just great for anybody who's
three inches tall.
So these are pretty limiting devices.
But it seems to me that they are going to be very important
because of the prevalence of these devices, especially in
areas where alternative access to the
internet isn't available.
What is interesting about these is that because you
carry them around on your person or in your purse, they
become your information access method,
your information source.
And often, you want information which is relevant
to where you are at the moment.
So geographically indexed information is becoming very,
very valuable and very important, especially as you
access it through mobiles.
I have a small anecdote to share which emphasized for me
the importance of having access to geographically
indexed information.
My family and I went on a vacation this May in a place
called Page, Arizona.
It's adjacent to something called Lake Powell.
We decided to rent a houseboat.
Believe me, if you like steering things around, don't
rent a houseboat.
It steers like a houseboat.
Anyway, I was terrified that I was just going to ricochet my
way down the lake.
But in any case, the problem is once you get on the boat,
there's no place to get any food.
So you have to prepare by buying food
and bringing it onboard.
So as we were driving into Lake Powell, we were
discussing what meals we were going to produce.
And somebody said, well, I want to make paella.
And I thought, well, that's interesting.
You need saffron to do that.
Where am I going to find saffron in this little town of
Page, Arizona?
So fortunately, I got a good GPRS signal.
So I pulled out the BlackBerry.
And I went to Google.
And I said, Page, Arizona grocery store saffron.
And up popped a response with an address, a telephone
number, name of the store, and a little map to show you how
to get there.
So I clicked on the telephone number.
And, of course, this being a telephone, it made the call.
The phone rang.
Somebody answered.
And I said, could I please speak to the spice department?
Now, this is a little store.
So it's probably the owner that, this is the spice
And I said, do you have any saffron?
He says, I don't know.
He went off.
And he came back.
He says, yeah, I've got some.
So we followed the map.
This all happening in real time.
We follow the map, drive into the parking lot.
And I ran in and bought $12.99 worth of saffron.
That's 0.06 ounces, in case you wondered.
And we went off on Lake Powell.
We made a really nice paella.
What really struck me is that I was able to get information
that was relevant to a specific need in real time and
execute this transaction.
It would not have worked--
can you imagine going and trying to find something in
the white pages or the yellow pages at a gas station
or what have you?
The fact that you can get information is useful to you
in real time is really quite striking.
So I believe that as this mobile revolution continues to
unfold, that geographically indexed information is going
to be extremely valuable.

Well, some of you have been around for a while and have
watched the internet grow.
I've been a little stunned at some of the devices that are
starting to show up on the network, like internet enabled
refrigerators or picture frames that download images
off of web sites and then cycle through them
automatically, or things that look like telephones, but
they're actually voice over IP computers.
But the guy that really stunned me is the fellow in
the middle.
He's from San Diego.
He made an internet enabled surfboard.
I guess he was sitting out on the water thinking, you know,
if I had a laptop in my surfboard, I could be surfing
the internet while I'm waiting to surf the Pacific Ocean.
So he built a laptop into his surf board.
And he put a WiFi service in the rescue
shack back on the beach.
And he now sells this is a product.
So if you're interested in buying internet enabled
surfboard, he's the guy to go to.
I honestly think that there are going to be billions of
devices on the net, more devices than there are people.
And if you think about the number of appliances that
serve you every day, there are lots of them.
And imagine that they're all online.
Imagine being able to interact with them or use intermediary
services to interact with them.
So as an example, instead of having to interact directly
with your entertainment systems, if they were up on
the network and accessible that way, you might interact
through a web page on a service on the network, which
then turns around and takes care of downloading movies
that you want to watch or music that you want to listen
to or moving content from one place to another.
All of that could be done through the internet.
In fact, a lot of those devices have remote controls.
And if you're like me, there are lots of them.
And then you fumble around trying to figure out which
remote control goes with which box.
And after you figure that out, that's the remote control with
the dead battery.
So the idea here is to replace all those with your mobile,
which is internet enabled.
So are the devices in the room.
You program them and interact with
them through the internet.
So you don't even have to be in the same room.
Gee, you don't even have to be in the house.
You could be anywhere in the world where you could get
access to the internet, and you could control your
entertainment systems.
Of course, so could the 15-year-old next door.
And so you clearly need strong authentication in order to
make sure only the authorized users of these systems are
controlling your entertainment system, or your heating and
ventilation, or your security.
All of these things could easily be online--
lots of appliances at home, appliances in the office, all
manageable through the internet and offering
opportunities for third parties to help manage some of
that equipment for you.
So this kind of the network of things is creating
opportunities for people to offer new
products and services.
I don't have time to go through all of
these various examples.
But there are little scenarios you can cook up, like in the
refrigerator that's online.
If your families are like American families, the
communication medium between American families is generally
paper and magnets on the front of the refrigerator.
And now, if you put up a nice laptop interface on the front
of the refrigerator door, you can communicate with family
members by blogging, by instant messaging, and by web
pages and email.
But it gets more interesting if you imagine that the
refrigerator has an RFID detector inside.
And RFID chips are on the products that you put inside
the refrigerator.
So now the refrigerator can know what it has inside.
And while you're at work, it's surfing the internet looking
for recipes that it knows it could make
with what it has inside.
So when you get home, you see a nice list of things to have
for dinner if you like.
And you can extrapolate this.
You might be on vacation, and you get an email.
It's from your refrigerator.
It says, I don't know how much yogurt is left, but you put it
in there three weeks ago, and it's going to
crawl out on its own.
Or maybe your mobile goes off.
It's an SMS from your refrigerator.
Don't forget the marinara sauce.
I have everything else I need for spaghetti dinner tonight.
Now, unfortunately, the Japanese have spoiled this
beautiful scenario.
They've invented an internet enabled bathroom scale.
When you step on the scale, it figures out which family
member you are based on your weight.
And it sends that information to the doctor to become part
of your medical record.
And, of course, that's OK, except for one problem.
The refrigerator's on the same network.
So when you come home, you see diet recipes
coming up on the display.
Or maybe it just refuses to open because it
knows you're a diet.
I'm going to skip over-- oh, wait a minute.
I'm sorry.
There's some important stuff here.
I mentioned earlier IPv6.
P And that's something that really is a major
transformation of the core part of the internet at the
internet protocol layer.
There are other things that are happening in 2007 and now
2008 that are going to have an impact on all of us who offer
various kinds of internet service.
One thing is the introduction of non-Latin top level
domains, internationalized domain names, that are written
in character sets that include things like Arabic, and
Cyrillic, and Hebrew, and Chinese of various kinds, and
Kanji, and Hangul, and so forth.
ICANN has already put up 11 test languages in the top
level domains, in the root zone file.
And it's encouraging people to go there and to try out
interactions with those domain names using various
application software packages, including browsers and also
email to give you a chance to see how the software will
interact with these non-Latin character domain names.
They're are represented typically in Unicode.
It may be Unicode encoded in UTF-8, for example.
But what's important is that the software has to recognize
that these are domain names, even though they're expressed
using strings other than simply A through Z and zero
through nine and a hyphen.
The other thing which is going on is that digital signing of
domain name entries.
So DNSSEC is a way of allowing someone who's doing the domain
name lookup to ask for a digitally signed answer.
And when that comes back, you have some validation that the
information in that domain name entry has not been
altered, that it has maintained integrity from the
time it was put in.
This is not encrypted anything.
Is simply a question of digitally signing things to
make sure that the information is as valid as it was when it
went in the first place.
Those things are all under way now in the internet.
And they may have an impact on every one of us that are
involved in building systems that run on the internet.
I'd like to just quickly go through a couple of other
points here.
One of them is that intellectual property handling
in the online environment is becoming quite a challenge.
Digitized information is easy to copy, and it's easy to
The philosophy behind copyright laws, including the
Bern Convention that was developed here in Switzerland,
says that physical copies of things are what we are
concerned about.
And the difficulty of copying or the cost of copying
physical objects is what has made that particular law
But in the presence of digital versions of these things, it's
turning out to be much harder to enforce.
It may very well be that we need to back away and rethink
what copyright means in this online and digital
There are alternatives that have been suggested.
Creative Commons is one of them, celebrating its fifth
anniversary this year, that may find alternative ways of
compensating authors or of letting authors say whether
they want to be compensated or in what way they want to be
compensated for their intellectual property that's
been put into this online environment.
Tim Berners-Lee has worked for some time and spoken often
about the semantic web.
This is an idea that allows us to interact with the content
of the network, not just with strings, but with some notion
of the meaning of the strings.
That project is still a work in progress.
If in fact it's possible to codify or otherwise indicate
the meaning of contents on the net, it would be very
beneficial, certainly, from Google's point of view.
Because today, we tend to navigate you to a document.
But what you really wanted was answers.
And in order to do a better job of helping you find
answers, we need to understand the semantics of
what's in the net.
And right now, we can't do that.
I'm becoming increasingly concerned about the nature of
the objects that are in the internet today.
Some of them are extremely complex.
There not interpretable without software.
A spreadsheet, for example, is a dead thing until you
actually bring it up in the spreadsheet program and
interact with it.
It's a very complex object sitting on the net somewhere.
You can't simply print it out.
Well, you can, but you get a very limited representation of
the real content, meaning and, complexity of the objects.
So I'm unconcerned that over time, the contents that we put
into the internet will be dependent on software to
interpret what the bits mean.
That leads me to my biggest worry, which
I'll call bit rot.
If in fact bits are stored away over time, and they're
moved from one medium to another as new storage media
come along, what will happen if we lose access to the
software that knows how to interpret the bits?
At that point, you won't know what you have other
than a bag of bits.
And so the question is, what we do about that?
Let me give you a scenario.
It's the year 3000.
And you've just gone through a Google search.
And yo've turned up a PowerPoint file from 1997.
So suppose you're running Windows 3000.
The question is, does Windows 3000 know how to interpret the
1997 PowerPoint?
And the chances are it does not.
And this is not arbitrary dig at Microsoft.
Even if this were open source software, the probability that
you would maintain backward compatibility for 1,000 years
strikes me as being fairly low.
So the question is what to do if you're thinking about
information that you want to be accessible 1,000 years from
now like vellum documents are today that
are 1,000 years old.
We have to start thinking about how to preserve software
that will be able to interpret the bits that we keep saving.
You may even have to go so far as to save the operating
system that knew how to run the application that could
interpret the bits.
And maybe even emulate hardware that ran the
operating system that knows how to run the application
that can interpret the bits.
There is very, very little progress made right now in
that domain, something that we should be
very concerned about.
Otherwise, 1,000 years from now, historians will wonder
what the heck we did in the 21st century.
There will be nothing about us other than a
pile of rotting bits.
And that's all they will know, is that we were the rotten bit
And I'm sure that's not what we want them to know about us.
OK, last update.
And I have a whole bunch of questions here that people
have asked already.
And I'll try to answer some of them.
This project that I'm about to tell you about is
not a Google project.
Google lets me have time to work on it.
But I don't want you to walk out of this auditorium saying,
hah, I've figured out what Google's business plan is.
It's going to take over the solar system.
That's not what this is about.
This is about supporting the exploration of the solar
system using standardized communication protocols.
Because historically, we have not standardized these
communication systems in the same way that we've
standardize communication in the internet.
Now, we all know we've been exploring Mars
using robotic equipment.
Usually, to communicate with the spacecraft, we use the
Deep Space Network, which was developed in 1964.
These are big 70-meter dishes in Goldstone, California,
Madrid, Spain, and Canberra, Australia.
There are also adjacent to 35-meter antennas as well.
Half of one is kind of visible over the right hand corner of
that image.
So as the earth rotates, these big deep space dishes are
rotating along and seeing out into the solar system, able to
communicate with spacecraft like this one, which may be in
orbit around a planet or flying past an asteroid, or in
some cases actually landing on the surface of a planet like
the rovers in 2004.
One thing that you might not know is that most of the
communication protocols that are used for these deep space
missions are tailored to the sensor platforms, the sensors
are on board the spacecraft platforms, in order to make
most efficient use of the available communication
capacity, which is often fairly limited.
The rovers that went onto Mars in the beginning of 2004 are
still running, which is pretty amazing, considering their
original mission time was only 90 days.
So they're still in operation, although one of them--
I forget which one-- has a broken wheel, and it's kind of
dragging furrows in the Martian soil.
But they're still operational.
One of the problems that showed up, though, very early
in the rover mission is that the plan was to transmit data
with the high gain antenna.

There's a thing that looks like a pie tin on
the right hand side.
That was the high gain antenna that was supposed to transmit
data straight back to the Deep Space Network from
the surface of Mars.
When they turned the radios on and started transmitting, they
And it happened on both spacecraft, so it
was a design problem.
So they had to reduce the duty cycle to avoid having the
radio damages itself, which really drove the principal
investigators crazy. because 28 kilobits wasn't very much
to begin with.
And now it's less frequent transmissions.
So the guys at JPL figured out a way to essentially
reconfigure systems so that the data could be transmitted
from the rover up to an orbiter.
And there were four orbiters available around Mars.
They were reprogrammed in order to take the data up on a
128 kilobit radio, a different radio, which didn't
have very far to go.
So the signal to noise ratio was high enough to get much
higher data rates.
Then the data was stored in the orbiter until it got
around to the point where it could transmit the data to the
Deep Space Net.
And once again, it could transmit at 128 kilobits a
second, partly because the orbiters had bigger solar
panels and more power available than the rovers on
the surface.
So the net result is that all the information that's coming
back from the rovers is going through a store and forward
system, which, of course, is the way the internet works.
This reconfirmed an idea that my colleagues at the Jet
Propulsion Lab and I have been pursuing since 1998.
And that's the definition of protocols to run an
interplanetary extension of the internet.
Basically, we assume we're running TCP/IP on the surface
of the planets and in the spacecraft.
Those are low latency environments.
TCP/IP works very well there.
We thought we could get away with running TCP/IP for the
interplanetary part.
That idea lasted about a week.
It's pretty obvious what the problem is.
When Earth and Mars are farthest apart in their
orbits, they're 235 million miles apart.
And it takes 20 minutes one way at the speed of light for
a signal to propagate.
And you can imagine how flow control will work with TCP.
You'd say, OK, I'm out of room now.
The guy at the other end doesn't hear you say that for
20 minutes.
Of course, he's transmitting like crazy.
And then packets are falling all over the place.
It doesn't work.
To made matters worse, there's this thing
called celestial motion.
The planets have this nasty habit of rotating.
And so you can imagine you're trying to talk to a rover on
the surface of Mars.
And after a while, it rotates out of sight and you can't
talk to it until it gets back around to the other side.
So the communication is disrupted.
So we concluded very quickly that we were faced with a
delay and disruption problem and needed to invent a set of
protocols that built that into its assumptions, which were
frankly not part of the internet assumptions, the
TCP/IP protocols.
So we developed a set of protocols.
We've been going through tests of them.
We've got to the point now where we have publicly
available software.
It's all up on the DTNRG website--
Delay and Disruption Tolerant Networking Research Group--
dot org.
And that protocol has now been tested in terrestrial
We picked two.
The Defense Department, DARPA, funded the original
interplanetary architecture work.
And after we got done and realized this DTN thing was a
serious problem, we realized that they had a problem in
tactical communication, the same problem, disruption,
delay, uncertain delay.
So we went back and said, we think you have a problem.
And we think you could use the DTN protocols for tactical
military communication.
So they said, OK, prove it.
And we said, OK.
We went off, and we built some--
these motes that come from Berkeley, the little Linux
operating systems. We built a bunch of motes and put the DTN
protocols on board.
And then we went to the Marine Corps.
And we said, OK, we'd like to test this stuff out.
What application would you like to run?
And they said, chat.
I said, are you kidding?
You're sitting here with bullets whizzing by and you're
going like this?
And they said yes, because chat has the nice feature that
when you reconnect, all the exchanges that took place are
then given to you that you missed.
And so you get back in sync with the other people who are
part of this communications environment.
So we said, well, OK.
So we did that.
We implemented it.
We went out to Northern Virginia to a test deployment
of these things with the Marine Corps.
And it worked.
And so I thought that was pretty cool, demonstration of
DTN working terrestrially, something useful
coming out of all this.
And the next thing I knew, they'd taken all of
our stuff to Iraq.
And I said, wait a minute, it's an experiment.
And they said, no it isn't.
And off they went.
So we said, all right, fine.
Then we thought, well, let's try this in a more civilian
environment as well.
Some of you, I'm sure, are familiar with the fact that
the reindeer herders, the Sami are in the northern part of
Sweden, and Finland, and Norway, and Russia.
And they are pretty isolated, because they're so far north.
And satellite communication is a problem because the dishes
are, bang, right there on the horizon.
So we said, well, what would happen if we stuck a laptop
with 802.11 in it and the DTN protocols
in all-terrain vehicle?
And we put WiFi service in the villages.
So we tried one village.
And we tried this random interaction with the
system using DTN.
And it worked.
That was last year.
So next year, we're going to try a multiple village test of
the DTN protocols to see whether works.
And if it works out well enough, then maybe we'll put
these things in the snow mobiles so that it'll work
both during the summer in the winter.
So we're very happy that we've got terrestrial examples of
the use of the DTN protocols.
We're now the point where we're ready to start space
based testing.
2009, we're hoping to put the DTN protocols on board the
International Space Station.
And in 2011, NASA has offered to allow us to put the
protocols on board the deep impact spacecraft that already
completed its primary, mission which was to launch a probe
into a comet and gather data back.
But it's still out there, and it's still functioning.
So somewhere around 2011, we hope to space
qualify the DTN protocols.
And after that, Adrian Hooke, who's my counterpart now at
NASA, is also Chairman of the Consultative Committee for
Space Data Systems. And we hope to introduce that as a
standard protocol for use in space communication.
So what we're expecting, if we're lucky, is that the space
agencies around the world will adopt this is a standard.
They'll use it for every mission that they launch.
And that means that every time you launch a new mission, any
previous mission assets that are available can become part
of the support structure for the newly launched mission.
What will happen as a result is it will accrete an
interplanetary backbone over a period of decades as more and
more of these missions get launched in the system.
So me let me stop there and thank you again for
taking all this time.
And we'll see whether we can answer a few questions here if
that's OK with you.
So let me--
these are questions that apparently were submitted by
many of you here.
And I'll answer a few.
And then we'll see if we get some immediate
ones from the floor.
First company says, for many companies, the breakdown of
the internet for, say, three days would lead to a
substantial damage.
Question, is it possible to estimate the probability of
such a breakdown of the entire internet?
And if so, how could it be down?
And are there any numbers available?
Or maybe how could it be done?
I think the answer is I'm not going to tell you
how it could be done.
The answer is that the probability that the entire
internet could be taken down seems to be pretty small.
There have been plenty of opportunities over the past 20
years or so since the 1983 rollout of internet to destroy
it in one way or another.
And in spite of the fact that denial service attacks are by
far one of the most serious threats to internet stability,
it seems unlikely that the entire internet
would be taken down.
I will observe, however, that we manage to shoot ourselves
in the foot fairly regularly by mis-configuring things.
So if you mis-configure the routing cables of the routers,
you can easily damage significant
parts of the internet.
And we seem to do that with more frequency
than we would like.
But I think on the whole, the robustness of the system has
been pretty substantial.
That doesn't mean we shouldn't be introducing increasing
amounts of security mechanisms into the network in order to
limit that risk.
What's the biggest fallacy about the internet?
Well, one of them is that Al Gore invented it.
He didn't invent it, but he did have
something to do with it.
The other big fallacy is some people think the internet
happened because a bunch of local area networks got
together one day and said, let's build a
multi-network system.
The fact is that it started with wide area networks and
took a long time.
Will there ever be secure web apps based on a browser alone?
Would apps be more secure if run in applets?
Boy, that's a really good question.
Right now, the most vulnerable part of the internet world is
the browser.
Browsers ingest Java code or other high level codes, and
they often are unable to detect that this code is
actually trying to take over the machine, or install a
Trojan horse, or do some other damaging thing.
If we collectively were to invest anything at all, I
think we should be investing in building much, much smarter
browsers that are able to defend against some of the
dangerous downloads.
Should the evolution of the basic protocols
such as HTTP go--
or where should it go in order to support more asynchronous
interactions of web 2.0?
And a related question, is there too
much overhead in HTTP?
Well, first of all, I'd say that we should not depend
solely on HTTP as the medium of interaction on the net.
Peer-to-peer applications are really interesting.
And although some of them get abused for copying and
distributing material that's copyright, in fact, they often
are a very efficient way of linking people who want to
communicate in the network, either peer-wise or in
multiple groups.
So my reaction right now to this one is that we really
should be looking at asynchronous, peer-to-peer
kinds of interactions in addition to the more classical
HTTP designs.
Can the world wide web help humanity to become a just,
democratic society?
Well, the short answer that is no.
Probably not, although to be really honest and fair, the
internet probably is the most democratic communication
system that we've ever had, because it allows so many
people to introduce content, and to share it,
and exchange it.
But humanity is what it is.
Shakespeare keeps telling us about them.
That's why the plays are still so interesting.
So for humanity to become a just and democratic society is
going to take a fair amount of adjustment of the human beings
who make up that society, not just the technology that
surrounds them.
Authentication on the internet typically means complete
How can the privacy of users be protected?
The answer is that we need to do both.
We need to have anonymous access to the internet.
And we also need to have the ability to do strong
And the reason that you want to do both is that sometimes
it's important to be anonymous.
We all understand you can abuse the anonymity, and do
bad things, and stalk people, and say things
that are not true.
But there are also times when it's important to be anonymous
in order to allow whistleblowing.

On the other side, there are transactions that we want to
or engage in for which we really do need to know who the
other party is.
And so we want strong mechanisms to allow people to
validate each other to themselves, or to validate the
website to you.
And the same time, we also have to support anonymity.
And I think we need both.
I'll tell you what.
There's a list of almost 23 three questions here.
And I have the feeling that there are probably people in
the audience who would like to ask some questions that they
didn't ask ahead of time.
So let me stop with the pre-asked questions and ask if
there's anybody who would like to ask a question
live from the floor.
There's a microphone down here.
And if there's a brave soul who wants to ask the question,
I promise I won't spit.
And if there aren't any, I'll be happy to either go on or
maybe run off the stage.
Let's see.

When do you think ownership of the top DNS will be
transferred from the US government to an international
organism such as the UN?
Well, I'll be honest and say I hope it doesn't get
transferred to an international
organism such as the UN.
I think that transfer of the ICANN operation to a
multilateral organization would politicize it.
I'd point out that ICANN is a multi-stakeholder
organization, which means that governments, the private
sector, the technical community, the general public
have access to the process of policymaking for the domain
name system and internet address allocation.
It would be much better for this process to end up in a
multi-stakeholder structure, not a multilateral structure
like the UN.
The Internet Governance Forum just finished its meetings in
Rio de Janeiro.
It too is a multi-stakeholder organization.
And the conversations that take place among those
different stakeholders, I think, are extraordinarily
illuminating when it comes to seeing what the different
perspectives are about policy.
So I hope that the answer is, first, it doesn't end up in a
multilateral group, but rather a multi-stakeholder one.
And I would suggest that there isn't very much more to be
done to extract the US government the role here.
All it does right now in the Department of Commerce is to
validate that ICANN has followed its procedures to do
delegations of top level domains.
And that's all it does.
It's never forced any decisions on ICANN.
It's never rejected any recommendations
that ICANN has made.
This isn't to say that people would like to see the US, many
of them, not have this special relationship.
And I hope that sometime, certainly, in 2008, there's an
opportunity to revisit that relationship when we review
what's now called the Joint Project Agreement between the
Department of Commerce and ICANN.
Well, let me stop there.
And if there aren't any questions, I think maybe we
can wrap up.
I don't know whether you would like to say a closing
I just would like to say thanks to Vint Cerf.