`
`I N T E R V I
`
`E W
`
`“THE INTERNET
`
`IS THE
`
`INFORMATION AGE.
`
`BUSINESS PEOPLE
`
`KNOW THAT, EVEN IF
`
`THEY DON’T HAVE
`
`A CLEAR IDEA
`
`OF WHAT THE
`
`BUSINESS MODEL IS.
`THE INTERNET IS
`
`THE HOPE OF THE
`
`FUTURE.”
`
`–BOB METCALFE
`
`6
`
`BOB
`METCALFE
`ON WHAT’S
`WRONG
`WITH THE
`INTERNET:
`IT’STHE
`ECONOMY,
`STUPID
`
`B ob Metcalfe, self-described technology pundit, eminently
`
`successful engineer-entrepreneur, and International Data
`Group Vice President and InfoWorld columnist is the sub-
`ject of this month’s Internet Computing interview. Metcalfe’s inven-
`tion of the Ethernet in the early 1970s grew out of work that had
`begun with his 1973 thesis at Harvard, Packet Communications
`(republished in 1996 by Peer-to-Peer Communications, San Jose,
`Calif.), and his study of the Alohanet, a radio packet communica-
`tions network created by Norman Abramson and Franklin Kuo at
`the University of Hawaii. With David Boggs, Metcalfe developed the
`first Ethernet interfaces in 1973 that led to the landmark paper,
`“Ethernet: Distributed Packet Switching for Local Computer
`Networks,” to a patent application in 1975, and ultimately to the
`formation of 3Com and the adoption of Ethernet as an IEEE 802
`standard in 1982.
`
`1089-7801/97/$10.00 ©1997 IEEE
`
`IEEE INTERNET COMPUTING
`
`
`
`S C A L
`
`I N G T H E
`
`
`
`I N T E R N E T
`
`.
`
`After leaving 3Com in 1990,
`Metcalfe has continued to serve the
`field of networking through his role
`as a journalist. Clearly a believer
`in the watchdog role of the press,
`Metcalfe has been in the media
`spotlight for the past year because
`of his flamboyant predictions of
`Internet “collapse.” In an interview
`with IC’s EIC Charles Petrie and
`staff editor Meredith Wiggins,
`Metcalfe said that people would
`read his columns and say, “What
`is he doing?” We hope this inter-
`view will answer that question.
`
`We spoke with Metcalfe on Feb-
`ruary 10 in Indian Wells at the
`DEMO conference sponsored by
`IDG. In preparation for the inter-
`view we had sent him a copy of
`our interview with George Gilder,
`contributing editor to Forbes and
`prominent author, which appeared
`in our first issue. Metcalfe began
`our interview by picking up
`Internet Computing and saying . . .
`
`Metcalfe: You know the more
`Gilder and I talk, the more I see
`we have so much in common.
`We’re both very conservative. And
`then technologically speaking we
`agree on almost everything. He
`makes a big point of my saying
`that we lack bandwidth, whereas
`he believes bandwidth is in abun-
`dance. Well bandwidth is in abundance and can be in
`abundance and will be in abundance—in time. It’s a ques-
`tion of time frame.
`
`The biggest point of disagreement that I’ve noticed
`between us—and I hasten to reemphasize it’s a very small
`disagreement—is on the issue of wireless technologies.
`He’s very gung ho about wireless, and I’m very pessimistic.
`
`©1997 Alan J. Duignan
`
`Especially in product magazines like InfoWorld, look and
`you’ll see that we’re already in a wireless world. (Points
`to an ad in InfoWorld.) Oh! there’s wireless. Even per-
`sonal computer products in ads are wireless already.
`Advertisers have figured out that people want their
`equipment wireless so they take the wires out of the pic-
`tures. Show me the wire. Here’s a picture of a PC for
`sale. There’s no wires in it.
`
`IC: Why are you pessimistic? There is certainly a lot of
`investment going on.
`
`IC: There’s not even a mouse wire. But evidence of wish-
`ful thinking still isn’t proof that wireless won’t work.
`
`Metcalfe: Because I’ve heard the story before. I mean,
`I’ve been to this ball game, and the main problem is we
`all desperately want everything to be wireless. We want
`it so bad that we’re willing to believe anything anyone
`tells us about wireless.
`
`Metcalfe: No. It explains why the story is eternally with
`us that wireless is going to solve all our problems. Every
`time a random guy shows up and says, “I’m gonna make
`you wireless,” we want it so desperately that we pump it
`up. And my conjecture is—and once again I’m talking
`
`IEEE INTERNET COMPUTING
`
`http://computer.org/internet/
`
`MARCH • APRIL 1997
`
`7
`
`
`
`.
`
`I N T E R V I
`
`E W
`
`USING THE REAL ETHER WILL
`ALWAYS BE MUCH MORE
`EXPENSIVE AND MUCH SLOWER.
`
`But seriously, what I’m saying is that in
`the future, computer networks (which
`is where I am expert, as opposed to
`cellular telephones where I’m a pitiful
`user) won’t be wireless. In fact I’ve
`attracted the ire of the wireless com-
`munity in saying that there will be
`wireless mobile computers, but they’ll
`be just like pipeless mobile bathrooms.
`I’m trying to make an analogy between
`bathrooms and computers. There’ll be
`as many mobile wireless computers as
`there are mobile, pipeless bathrooms.
`There are mobile pipeless bathrooms
`in airplanes, in ships, construction
`sites, sporting events. But in fact most
`of the networking in the world will be
`like bathrooms are. There’ll be pipes.
`
`IC: What’s the basis of your prediction?
`
`©1997 Alan J. Duignan
`
`Metcalfe: Well, take satellite. I’ve
`recently been out to visit Hughes,
`who has DirecPC, which is wireless
`satellite distribution of data. It’s a very
`exciting product; I’m rooting for it.
`And they have a new follow-on prod-
`uct which they call SpacewayW which
`I’m very enthusiastic about. But they
`need help in encouraging themselves to make the $3 bil-
`lion investment to put up these eight satellites. And
`when it’s all said and done, the amount of bandwidth
`that would be provided by such a system is a drop in
`the bucket. It’s a few gigabits per second. One optical
`fiber strung from here to New York would provide much
`more bandwidth. To satisfy the Internet’s needs for
`bandwidth you’d have to blacken the sky with satellites.
`
`about a one percent disagreement—that George Gilder
`has been suckered because, like all of us, he wants it to
`be wireless. And while the people who are peddling
`wireless technology are sincere in their efforts, they’re
`exaggerating its effectiveness. Oh, there will be wireless,
`there is wireless. Look what I have on my belt. I have
`this wonderful wireless device. (Produces a cellular
`phone.) I’m trying to make it a prosthetic, the way my
`glasses are, so that it’s always with me.
`
`See my glasses are always with me, and my pen, and I’m
`trying to be this wireless guy. But really, this StarTAC is gor-
`geous and wonderful. Of course, it doesn’t work all the
`time. If you open it up, a large part of this device is dedicat-
`ed to proving that it doesn’t work. See these lights? These
`lights are important; they tell you if it’s working now.
`
`And then there’s the battery. The battery doesn’t really
`work. So there’s an enormous number of lights dedicated
`to proving that the battery doesn’t really work. It doesn’t
`work so often that you have to be constantly aware that
`it’s about to run out.
`
`W URLs from this page
`Spaceway • www.hcisat.com/SPACEWAY/revolution.html
`
`MARCH • APRIL 1997
`
`http://computer.org/internet/
`
`8
`
`If you do the arithmetic, the satellites won’t “win” against
`optical fibers, because the capacities are orders of magni-
`tude out of whack. That doesn’t mean they have to win.
`We’re going to see satellites being very useful for broad-
`cast applications, for highly mobile applications, and for
`highly remote applications. But you’re not going to elim-
`inate optical fibers with satellites.
`
`IC: Will wireless then need a different pricing structure?
`
`Metcalfe: Yes. The simple fact is that wireless uses one
`copy of the ether. There’s only one copy of it, and they
`all have to use it, and eventually they’ll run out. Whereas
`each optical fiber is its own copy of the ether; when you
`run another one, you have a whole new spectrum. You
`can duplicate the ether.
`
`IEEE INTERNET COMPUTING
`
`
`
`S C A L
`
`I N G T H E
`
`
`
`I N T E R N E T
`
`.
`
`done, and the customer support, and all the telephone
`poles. So dark fiber is a joke.
`
`So George is right, and I agree with him. As I agree with
`him on most other things. There will be abundant band-
`width, but it’s all dark silicon. It’s just so far from fruition.
`
`What’s going to take it to fruition? We need an economic
`model for that. And Gilder and I would agree again that
`the way to do it is through free market processes, invest-
`ment capital, and the technological advance fed thereby.
`
`Now I figured out recently that I agree with what Al Gore
`says about most things related to the Internet, which is a
`big surprise because he’s a Democrat. But for a long time
`he’s been saying words that I agree with: that the Internet is
`not going to be built by government. It’s going to be built
`by private industry. Now we’re finding out what he means:
`Government regulation is going to force private industry to
`build the Internet. He’s recently come out in favor, and the
`FCC and Reed Hundt have recently come out in favor, of
`
`Now wireless people by cellularizing are hoping to reuse
`the ether, and there’s promise in this. And this is where
`George Gilder might turn out to be right, although I
`don’t think so. If they cellularize down to cells that are
`really small, like this room here, then they get to reuse
`the ether; they’re making copies of the ether in a geo-
`graphical way. And now we’re slipping into intuition: My
`intuition is that using the real ether as opposed to copies
`of the ether in coax and in copper and optical fibers will
`always be much more expensive and much slower.
`
`These guys brag that they’re now running at 9.6 or 19.2
`kilobits per second. In the LAN world, where I was
`raised, 10 megabits per second was hot stuff. Now 100
`megabits per second is de rigueur, and gigabit Ethernet is
`coming. Those numbers are astronomically higher than
`19.2 kilobits per second, which these guys think is the
`greatest thing that ever happened. I’m sorry, there will
`always be a huge disparity, which means that we should
`not plan on the world being entirely wireless. It will have
`a mix of wireless and wired, but the predominance of
`data transmission will be optical fibers.
`
`IC: Do you agree with Gilder that transmission times
`through optical fiber are going to be radically improved
`using techniques like wavelength-division multiplexing?
`Is this another reason you don’t think wireless will win?
`
`Metcalfe: There’s great progress in optical fibers. Just
`speeding everything up with digital is great progress,
`then you have multiplexing, you have solitons and
`doped fiber and it’s gonna be great. Of course transmis-
`sion is not the whole game. There’s this other thing
`called switching. One of the funny phrases I laugh
`about is “dark fiber.” Telephone companies talk about
`dark fiber and the Internet people talk about dark fiber.
`By that they mean this fiber is all over the world, it’s
`everywhere. It’s just not being used. It’s excess capaci-
`ty. And were it not for the fact that they haven’t con-
`nected up the lasing diodes and the switching systems,
`all that bandwidth would be there.
`
`This is a little bit like—now I’m going to make a joke—
`it’s a little bit like talking about dark silicon. I mean
`there’s silicon all over the place! Look at Saudi Arabia,
`look at the desert. It’s dark silicon! It’s right out there!
`So Pentium chips and Power PCs and optical fibers are
`made of silicon. And it’s just all over the place! The
`trouble is, there’s a long way from dark to useful. Now
`I admit, a dark fiber that’s been installed is a long way
`from dark silicon, but it has that same sort of futile
`pregnancy about it. I know there’s a lot of dark fiber,
`but look at how much work has to get done to convert
`it into real bandwidth. All the switching has to get
`
`IEEE INTERNET COMPUTING
`
`
`
`.
`
`I N T E R V I
`
`E W
`
`forcing Internet service providers to provide Internet access
`to schools and libraries. Forcing them to give discounts to
`schools and libraries. That’s not free enterprise.
`
`IC: And the telcos are encouraging this because it pro-
`tects their monopoly.
`
`Metcalfe: You’re right, the
`telcos love this idea be-
`cause it drags the Internet
`under this regulatory um-
`brella that they know how
`to manipulate. It’s a terrible
`idea. Now the ISPs will
`begin to be reimbursed out
`of the Universal Service
`Fund, that artifact of the
`outmoded and discredited
`regulatory regime that the
`telcos flourish under. It’s
`like a bear hug for the
`Internet under the guise of
`schools and libraries, and
`gee, it’s so hard to say I
`don’t want schools and
`libraries connected to the
`Internet.
`
`IC: There’s a lot of contro-
`versy about universal ser-
`vice right now.
`
`Metcalfe: That’s right. The
`term universal serviceW got
`invented decades ago to
`describe the deal that we
`make through the federal government with the regulated
`monopolies. That is, in return for universal service we
`give them a monopoly, and then out of the monopoly
`profits that they’re able to make, we insist that they cross-
`subsidize. Urban subsidizes rural, rich subsidizes poor,
`business subsidizes residential. It’s a deal, and it’s a deal
`that basically has not worked, as the Internet has
`revealed. Here we are now with computers millions of
`times faster than they were recently, but the bandwidth
`isn’t there. The digital services aren’t there. ISDN is barely
`there, and it’s too expensive. And what I love to do is talk
`about this in terms of Moore’s law and Grove’s law. What
`
`W URLs from this page
`The FCC on universal service:
`• www.fcc.gov/learnnet/anhome.html
`• www.fcc.gov/Bureaus/Common_Carrier/Reports/decision.html
`• www.fcc.gov/Bureaus/Common_Carrier/News_Releases/1996/
`nrcc6077.html
`See also the Merit Network • www.merit.edu/k12.michigan/usr
`
`MARCH • APRIL 1997
`
`http://computer.org/internet/
`
`10
`
`Moore’s law says in essence is that microcomputers get
`twice as good every 18 months. Grove’s law (Intel’s CEO
`Andy Grove) on the other hand, says that bandwidth dou-
`bles every 100 years. And the reason it doubles every 100
`years is because we have this malfunctioning, underper-
`forming regulatory regime,
`that most people agree now
`is malfunctioning and that’s
`why we have the Tele-
`communication Act of 1996.
`
`IC: Which contradicts Gil-
`der’s prediction that band-
`width actually increases
`faster than computer speed.
`
`Metcalfe: But we’re confus-
`ing what we’re talking about
`here. Gilder, when he says
`that, is talking about techno-
`logical advance. When
`Grove and I talk about it,
`we’re talking about what’s
`available, what’s deployed,
`what you can buy. And so
`there is in the lab all this
`technological advance, but
`you can’t buy it. It’s not for
`sale because this regulatory
`environment
`refuses
`to
`invest in deploying it. So we
`didn’t get ISDN during the
`1980s. We barely have it
`now, and it’s expensive, and
`all that’s being done under
`government supervision.
`
`©1997 Alan J. Duignan
`
`IC: ISDN did get deployed in Europe, where there’s even
`more regulation of the telecommunications industry.
`
`Metcalfe: And would you like to compare telephone
`rates in Europe with telephone rates here?
`
`IC: Well, that’s a good point. . . .
`
`IC: Most people at home use their modems over voice
`lines to connect to the Internet. We hear it said that the
`local telcos are very slow in responding to the new
`usage patterns this is creating as local connections are
`left open for extended periods.
`
`Metcalfe: Well first of all, Internet connection should
`not be going through the dial-up network. That’s a
`basic problem, and it needs to be fixed. The reason it’s
`
`IEEE INTERNET COMPUTING
`
`PEOPLE SHOULD PAY
`FOR WHAT THEY USE.
`IT’S A BASIC PRINCIPLE
`OF MARKET ECONOMICS.
`
`
`
`S C A L
`
`I N G T H E
`
`
`
`I N T E R N E T
`
`.
`
`THE TELCOS HAVE NOT INVESTED PROPERLY IN DIGITAL
`SERVICES, SO WE HAVE A 100-YEAR-OLD VOICE SYSTEM THAT
`WE INVENT THESE KLUDGY MODEMS TO GET THROUGH.
`
`going through the dial-up network is that the telcos
`have not invested properly in digital services, so we
`have a 100-year-old voice system that we invent these
`kludgy modems to get through.
`
`We don’t want the local telephone companies investing
`in additional voice-switching capacity to carry addition-
`al Internet lines. That would be a stupid waste of
`money. And trying to get the ISPs to pay access
`charges, to contribute to the Universal Service Fund, to
`boost the investment in voice services to carry Internet,
`is futile. Now the telephone companies would love this,
`because once again it would bring the ISPs under their
`regulatory regime. And these telephone companies own
`the 51 Public Utilities Commissions. They can’t wait for
`the FCC to get out of this, and then they can just do
`battle with the ISPs at the PUCs—to use a lot of initials.
`No, no, no, no. We don’t want this.
`
`IC: Is that what’s happening?
`
`Metcalfe: What’s stopping it is the ruling that the FCC just
`made that says if you’re a local telephone company, you
`must provide access to your cable plant to anybody who
`wants it, so that ISPs can bypass all that voice-switching
`stuff and connect directly to all those cables with their
`routers and IP stuff. That’s what should happen. And
`that’s what the law was supposed to enable, and that’s
`what Reed Hundt asked to have happen, and that’s what
`the telephone companies are apparently fighting tooth
`and nail.
`
`IC: So what you’re saying is there really is no free mar-
`ket. It’s all a matter of lobbying and laws and who can
`get there first.
`
`Metcalfe: Right. I think the PUCs and the FCC need to be
`bulldozed. The minor alterations that we’re making to
`those regulatory structures are not going to get us any-
`where, because they’re gonna fight us in the courts for
`years and years and years. And they’re going to use the
`schools and libraries as little pawns in this game. I’m so
`depressed by the subject.
`
`IC: Do you think that in the long run we will end up
`
`changing our pricing structure for local calls and charge
`by the call as they do in Europe?
`
`Metcalfe: I think the local flat-rate telephone service
`which has been mandated by the universal service
`regime—the PUCs and the FCC—is a bad idea, because
`basically it’s not reflective of cost. People should pay for
`what they use. It’s a basic principle of market economics.
`People should pay for what they use, and they should be
`able to choose from competing choices, and that will
`drive prices down and keep them viable.
`
`You don’t want a government body deciding what the
`prices should be, you want competition. Can you imag-
`ine if a government body regulated prices in the comput-
`er industry? You’d still be paying millions of dollars per
`month for your computers, because that would be “fair.”
`And the computer companies would all be in the busi-
`ness of proving how expensive computers are so that
`they could justify charging us lots of money for them,
`and so the government would let them do that. The
`more they spent, the more profitable it would be
`because the government would give them a fair return
`on higher costs. What we need is competitive services in
`telecommunications, and those prices will start going
`down. Gilder’s right; the technology to drive those prices
`through the floor is there, it’s just not getting through
`because it’s not in the interest of the regulated monopo-
`lies to drive the prices through the floor.
`
`IC: What needs to happen in terms of economic models
`to the Internet itself?
`
`Metcalfe: Well as a beginning step we need measurement,
`management, and money. The three M’s. First, there’s
`nobody measuring it. Management also needs to be added,
`but the Internet intelligentsia is opposed to management.
`They love this idea of anarchy, and that the Internet is bio-
`logical. These are the people who say in horror, “You want
`to centrally manage the Internet?” I say, well, I don’t see
`that we’re in danger at the moment of overmanaging the
`Internet. When we get to that point I’ll join you on that
`side. And money. Everyone wants it to be free or flat rate
`or as close to free as they can get. So they won’t move to
`fix the Internet in these three important ways.
`
`IEEE INTERNET COMPUTING
`
`http://computer.org/internet/
`
`MARCH • APRIL 1997
`
`11
`
`
`
`.
`
`I N T E R V I
`
`E W
`
`day. And during busy periods they
`were routinely hitting 30 percent.
`
`Now TCP was not designed for 30 per-
`cent losses. It’s robust and it keeps
`working, but it’s not optimized for that,
`and they are now modifying it to
`accommodate higher rates of packet
`loss, which is good. For example, the
`Internet Engineering Task Force has
`something called selective ACK. With
`TCP if you send 10 packets and one
`gets lost, you have to retransmit all 10.
`That multiplies errors. When you then
`think about 30 percent packet loss,
`suddenly it gets, whoa, big! Selective
`ACK is a way for TCP to say, “Well I
`got this one and this one, but I didn’t
`get that one. Could you please send
`me that one?”
`
`©1997 Alan J. Duignan
`
`I DON’T SEE THAT WE’RE IN
`DANGER AT THE MOMENT OF
`OVER-MANAGING THE INTERNET.
`
`IC: Talk to us about measurements.
`
`Metcalfe: The first question is, “So where are the mea-
`surements?” Let’s go find the measurements. Where
`would we find measurements on the operational perfor-
`mance of the Internet? I know where we’d go. We’d go
`to the North American Network Operators Group,W
`where we should find a hotbed of measurement activi-
`ty. And there is measurement activity there, but it’s very
`controversial. In particular at all the NANOG meetings,
`the big ISPs show up, and they say, “Those measure-
`ments are no good and those measurements are no
`good, and we refuse to participate in those measure-
`ments because they contradict our measurements.”
`What are your measurements? “I can’t tell you our mea-
`surements. That’s proprietary.” The secret measure-
`ments!
`
`I recently wrote a column about NetNow,W a measure-
`ment project out of the Merit NetworkW by which they
`measure packet losses and delays through various
`access points in the network and publish them on the
`Net. The trend is horrible. According to NetNow, the
`packet losses were averaging 2–4 percent in a 24-hour
`
`W URLs from this page
`North American Network Operator’s Group • www.nanog.org
`NetNow • nic.merit.edu/~ipma/netnow/
`Merit Network • www.merit.edu
`
`MARCH • APRIL 1997
`
`http://computer.org/internet/
`
`12
`
`But anyway, the people at NetNow
`put text up on their site saying, in
`essence, “We’re now losing 30 percent
`during our peak hours and some of
`this is due to the perverse engineering
`by the ISPs,” a very damning statement.
`
`So I sent out a sort of RFP. I took the quote that I just
`summarized for you, and I e-mailed it to NANOG and a
`bunch of other people I know and I said, “What do you
`think about this? What’s going on here?” Now the ISPs
`uniformly said, “This data is wrong.” In particular, one
`very famous ISP guy came and said, “This data’s bogus.
`It uses pings, it measures round trips, and it’s inconsis-
`tent with our data. You should throw it out. Forget it,
`don’t worry, this is just more of your collapse hysteria.
`Leave us alone, go away, you’re clueless.”
`
`Well, I checked with the NetNow people and they don’t
`use ping data, and it wasn’t two-way, and—not only
`that—this ISP wouldn’t give me their data that contradicts
`the public data. So I say, something’s gotta give, either
`the data gets fixed or the network gets fixed.
`
`IC: It’s normal, though, for businesses to have proprietary
`data that they withhold from the public.
`
`Metcalfe: It is detrimental to the health of the Internet
`not to have a higher degree of cooperation among the
`ISPs. The analogy I make is with the airline industry: It’s
`as though each airline had its own airports, airlines
`investigated their own crashes, and the industry found it
`profitable to run ads that said, “Fly our airline. We crash
`
`IEEE INTERNET COMPUTING
`
`
`
`S C A L
`
`I N G T H E
`
`
`
`I N T E R N E T
`
`.
`
`less than our competition.” This is the state of the
`Internet industry. It’s crude. It needs to be refined. And
`we need a higher degree of cooperation.
`
`IC: What did the NANOG people say about the NetNow
`data?
`
`Metcalfe: Something like 30 of them responded because
`they’re on that NANOG list every day. Most of them hate
`my guts. But they all responded, and there was no con-
`sistency in their response; they do not agree. Some
`blamed it on slow servers, some blamed it on routing sta-
`bility, some blamed it on lack of capacity, some blamed
`it on route computation, and some agreed with me that it
`was lost packets. They were all over the map. Some said
`it was the speed of the circuits going in and out. They
`have 100 different explanations. We need to measure it.
`I’m not saying I have the measurements; it just seems
`that no one knows. No one knows. And the people who
`are in the best position to know are not talking for nor-
`mal commercial reasons. MCI, Sprint, they have data.
`
`Metcalfe: Yes, I now focus more on people who have
`not made up their minds. I try to explain what I’ve found
`out. The experts on both sides of an issue are generally
`beyond help.
`
`IC: What was driving you when you made your collapse
`predictions?
`
`Metcalfe: Well I’m a journalist, a pundit is the right word,
`maybe. I’m not sure if that’s a kind of journalist, but I’m
`a technology pundit. Now I have a substantial advantage
`over my colleagues who work for this and other newspa-
`pers in that I’m also a technologist and I know a lot of
`what’s going on here. I also have another substantial
`advantage in that I’m 20 years older than they are and I
`can’t be bullied as easily as they can be. So, in regard to
`collapses, (a) I believe I’m onto something, and (b) my
`highest goal is to serve my readers.
`
`My readers use the Internet, they are gung ho on the
`Internet. They’re buying Internet services, they’re looking
`
`IC: Let’s talk about your prediction that
`the Internet would collapse in 1996.
`
`Metcalfe: Well, it’s interesting. The
`people who always thought I was a
`jerk think I’m really a jerk now
`because the Internet didn’t collapse.
`And there are other people who think
`it did collapse and that I was com-
`pletely right. I’m discovering again
`anew that what people look for in the
`press is confirmation, not truth. They
`look, and if they see something that
`they disagree with they tend to dis-
`count it. If they see something that
`they agree with, they tend to adopt it
`as truth.
`
`What I’ve learned about my high-pro-
`file collapse theories of last year is the
`people who didn’t think there was
`going to be a collapse last year didn’t
`think one happened, and the ones
`who were inclined to take the warn-
`ing and believe that there might be
`trouble ahead, think that horrible
`things happened to the Internet, and
`that I was somehow prophetic. And
`so nothing changed.
`
`IC: Has this changed the way you
`think about your role as a journalist?
`
`IEEE INTERNET COMPUTING
`
`
`
`.
`
`I N T E R V I
`
`E W
`
`I ALMOST REGRET CHOOSING THE WORD COLLAPSE,
`BECAUSE I’M A BIG FAN OF THE INTERNET.
`I’VE NEVER MEANT TO SUGGEST THAT IT WAS GOING TO DIE.
`
`for advice. My column is now number two in this maga-
`zine. These readers want to know about the Internet. So
`what drives me is being of service to them.
`
`InfoWorld’s business is evaluating vendors for buyers. We
`look at products and we evaluate them and we help our
`readers buy. ISPs are a kind of vendor, and I’m looking
`at them on behalf of my readers. But the fact is I’m also
`a former vendor and a former Internet person. So in
`addition I’m also kind of trying, in my own way, to help
`the ISPs get their act together. And I believe their act is
`not together.
`
`IC: So you’re making these predictions both as an engi-
`neer and as a journalist.
`
`Metcalfe: I think the Internet is tragically flawed. I’m try-
`ing to accelerate its being fixed. I’ve identified a series of
`problems that I’m working on. So I’m trying to provoke,
`and to advise the process for fixing the Internet because
`I’m very optimistic about it, and my readers are depend-
`ing on it more.
`
`IC: You are deliberately being controversial, trying to fan
`the flames.
`
`Metcalfe: I am a journalist now, I’m not an engineer. My
`job is to provide food for thought, and food tastes better
`when it’s spiced. And so I spice it up to make it interest-
`ing to people and provocative, and to stimulate discus-
`sion, and I get it. And I get readership too, and I’m not
`ashamed of that. In fact, now one of the metrics of suc-
`cess for me is to build readership. I’m happy when the
`readership surveys come in and they say that my reader-
`ship is up. It’s been going up for six years, and I just
`passed a half million readers a week. But that doesn’t
`negate my other aims.
`
`IC: Let’s turn to the problems that you’ve identified in the
`Internet. In December 1995, you predicted a spectacular,
`supernova collapse, much bigger than the 50,000 lines
`down for over an hour that the FCC requires to be reported.
`
`Metcalfe: What happened was that in December 1995 I
`wrote a largely tongue-in-cheek, sarcastic column,W
`which you can still read because it’s on the Internet.
`Then I began to refine this prediction and escape the sar-
`casm and funniness. A lot of people asked me what
`would constitute a collapse, and so I began to quantify it,
`and I used this 50K number, which happens to be an
`FCC guideline on telephone outages. That’s a 50 kilo-
`lapse: 50,000 users for an hour.
`
`Then in August it started happening. Netcom, in August
`of 1996, suffered a 5.2 megalapse: 400,000 users were
`denied Internet access for 13 hours. And then on August
`7th America Online suffered a 118 megalapse. That is, its
`then 6.2 million users lost Internet access for 19 hours.
`And it was at that point, quite late in the year, that I went
`out on a really long limb and said “This is just the begin-
`ning. We’re gonna see a gigalapse this year.” And that in
`fact is the prediction that did not come true.
`
`IC: Your original prediction was the spectacular superno-
`va collapse, which was unquantified. Do you think we
`had a spectacular supernova collapse?
`
`Metcalfe: No, I would tell you that in my meaning, the
`spectacular supernova collapse of the Internet—and by
`that I don’t mean AOL—did not occur. The Internet,
`however, bogged steadily down and it continues to bog
`down. It’s getting worse every day, and it’s conceivable
`that the spectacular supernova collapse of the backbone
`will occur.
`
`But there’s another point here. I changed my prediction
`during the course of the year from singular collapse to
`plural collapse, because too many people began to
`attribute to me that I was predicting the “death” of the
`Internet, which I don’t believe I ever predicted. In my
`mind it was going to suffer catastrophic collapses, recov-
`er, collapse again, recover, collapse again, until it got
`fixed. I almost regret choosing the word collapse,
`because I’m a big fan of the Internet. I’ve never meant to
`suggest that it was going to die.
`
`W URLs from this page
`See www.infoworld.com/cgi-bin/displayArchives.pl?dt_IWE49-95_23.htm
`
`I think if I had played it differently and suggested there
`would be a 100 megalapse in 1996, most people would
`
`MARCH • APRIL 1997
`
`http://computer.org/internet/
`
`14
`
`IEEE INTERNET COMPUTING
`
`
`
`S C A L
`
`I N G T H E
`
`
`
`I N T E R N E T
`
`.
`
`download than domestic sites, and download times from
`Europe to the US are even slower. Is anyone studying
`router policies to see if they advantage some people and
`disadvantage others?
`
`Metcalfe: Oh yes. They clearly do. When traffic comes
`into an ISP router, there’s this thing called “fastest exit
`routing” or “hot potato routing,” by which they say, “I’m
`not going to send this traffic by the shortest possible
`route to its destination because then it would travel on
`my circuits. Since I don’t get any revenue for this, I want
`
`have agreed that was horrible enough. The August AOL
`outage was 118. And that was a router collapse, by the
`way, not a front-end problem.
`
`IC: Many of the technical people are saying even if one
`site has a huge problem it will not propagate through the
`Net, and so far none has done so.
`
`Metcalfe: Well, wait a second. In the case of Netcom it
`propagated through all of Netcom, from Cisco router to
`Cisco router to Cisco router, and again in the AOL case it
`propagated through all their Cisco
`routers. We were fortunate, lucky is
`the way I would phrase it, that it did
`not propagate through the Netcom
`boundary. And, by the way, I don’t
`think that the users of AOL are unim-
`portant. It matters if they lose access.
`
`IC: Why do you believe we were lucky?
`
`Metcalfe: Because I know from talking
`to experts that the routing configuration
`process is ad hoc, highly complicated,
`and Cisco is working very hard to auto-
`mate, simulate, and protect that process.
`
`There are many kinds of catastrophic
`collapse, and some of them not traffic-
`related. In fact, the ones I’m most wor-
`ried about are not
`traffic related;
`they’re fragility related. Bugs in code.
`The DNS (domain naming system). If
`the nine DNS servers were somehow
`put out, the whole Internet would be
`gone. Packets would still be able to
`flow, you just wouldn’t be able to
`access anything because you wouldn’t
`know