Carl Malamud: Internet Talk Radio, flame of the Internet.
This is Carl Malamud. Welcome to Geek of the Week. We’re talking today to Brian Carpenter, who’s the group leader for the Communication Systems Group at CERN. CERN is the European laboratory for particle physics. Welcome to Geek of the Week, Brian.
Brian Carpenter: Well hello.
Malamud: I guess I’d first like to ask you about CERN’s role in the Internet. CERN is one of the kind of magnets. It provides connectivity to much of the world. Why is a physics laboratory so influential in providing general connectivity?
Carpenter: I think it’s because our main goal is to support remote users. Our main users are physicists who have to work most of the year in their own university departments. They have to teach their students. And once in awhile they would like to do some research. And the experiments have to be done in a central location, which in Europe happens to be at CERN, which is near Geneva. So, when these people are back home, they need access to their data, they need access to the experiments. If the experiment is running they want to be able to log in and check things from home. And that’s why it just became a priority to us to have network connectivity. And the best way to get connectivity was to be a hub.
Malamud: And you were a hub at first for HEPnet, the High-Energy Physics Network, which was based on DECnet.
Carpenter: Well historically I would say that that’s probably the first thing. You know, the Italian physics community really started using DECnet in Europe, and then it migrated to CERN, and grew outwards from CERN. And more or less at the same time we became one of the main nodes in the EARN network, which was and still is the European BITNET. So those two bits of connectivity grew up in the early 80s, I would say. Early to mid-80s. And that’s really how we got into networking as opposed to a few point-to-point links to external laboratories.
Malamud: And that networking in the HEPnet world was DECnet. You were one of the larger DECnet Phase IV sites, I believe?
Carpenter: Yes. It was a combination of DECnet and X25—X dot 25 in America, used as both as a base protocol and as a direct support for remote login. So that was a historical situation. And when the BITNET technology arrived, we started running RSCS some of the lines as well. So we’ve always been a multi-protocol site, but for a long time DECnet was probably the highest bitrate, if you like.
Malamud: Are you still a large DECnet site?
Carpenter: We’re still very dependent on DECnet because the experiments that’re now running, these are experiments were designed in the early 80s. They committed to computers from Digital Equipment Corporation; this is not a commercial, it’s a statement of fact.
Malamud: Physicists have often used VAXes, VAXes and Fortran.
Carpenter: Absolutely. VAXes and Fortran. And so these people found DECnet just came in the box with their computer, so they didn’t have too much choice. And that’s how we got into DECnet. And we’re clearly going to stay with DECnet as long as our users say they need it. Because we’re user-driven.
Malamud: Are you running DECnet Phase V or Phave IV?
Carpenter: We’re running both. As you probably know, some components of of DECnet Phase V are released. And we’re running them. We’re planning to migrate backbone routing to DECnet Phase V next month—that’s August. And not all the sites are going to migrate. This is a partial migration of the backbone. And you know, we have to do this because this is the current product, and you have to follow the current product. So you know, we think we’re going to be with DECnet for quite a while.
Malamud: Do you also run TCP/IP services?
Carpenter: I would say we do run a few TCP/IP services. This has grown at incredible speed. I mean, we started running TCP/IP on the LAN in the 1980s. And at a certain point, it became obvious that the Internet was going to grow in Europe. And this was a sort of spontaneous growth, and it became pretty obvious that we were not going to be able to deny our users this service. So at a certain point we got a collective decision in the HEPnet community to start supporting TCP/IP services. And having taken that decision, which was a political decision, we then started implementing it. Which meant buying routers, getting to know the technologies, setting up nameservers, and putting IP traffic in parallel to DECnet on quite a few of our leased lines, using multiplexes, while we got the experience. What’s happened is the growth rate in IP traffic has been quite fantastic, as the growth rate in a number of Unix systems has been fantastic. I think those two things clearly go together. And now it’s dominating our traffic by a large margin.
Malamud: In Europe, there has been historically a strong pressure to move towards OSI protocols. The DECnet Phase V world is a modification or implementation of the OSI protocols yet you’re also running TCP/IP. Do you feel caught in the middle there? Can you run a multi-protocol net effectively?
Carpenter: Well this is a very embarrassing question for me because I took over the group in the beginning of 1985, and I remember very well in May 1985 I came to a conference in Amsterdam, where we are today, and I gave a talk on the OSI networking policy of CERN. And I had to give this talk several times because I was new in the job and my bosses wanted to know what the new guy was going to do. So I got quite a reputation for talking about OSI.
Now, you know, it’s very difficult when you find egg on your face to wipe it off and still look sensible. So, Carl is smiling because I was wiping egg off my face in front of the microphone. It’s a real shame we don’t have video as well as audio.
Malamud: We’re working on that.
Carpenter: Okay. That’s good.
So you know, by ’88 or so it became obvious that OSI was just not coming fast enough, if it was coming at all. But, there was a lot of politics around that in Europe. There was some people who very honestly believed that OSI was going to be the solution to all our problems, and who didn’t modify that view as quickly as I did in the light of events, with the way that the Internet Protocol suite became available on pretty well everything you could buy and OSI was still theory.
So we went through a period which was politically quite delicate, I would say, of persuading our management that actually we’d been wrong by telling them OSI was the future; telling them IP was the future, at least for a while. And we can talk about how long that is. And so this is a situation we got into it.
Now, we’ve always continued some effort on OSI at CERN. We’ve participated in the RARE CLNS pilot. That’s the pilot project for using the OSI connectionless level 3. And in fact that’s still running; it’s even supporting part of the demonstrations here at the IETF. And you know, OSI has not gone away, but it hasn’t arrived either.
Malamud: It is going to arrive?
Carpenter: I would like to quote an official of the European Commission who showed a transparency once which said, “OSI has a constant property of being just around the corner.” And I think that’s absolutely true. You know, we’ve just been waiting too long. And I think this is why people’s patience in Europe snapped and everybody, with very few exceptions at a certain point realized that to deliver services to the users, there’s no alternative to the Internet protocol suite.
Malamud: Is OSI going to happen, or has it lost its window of opportunity?
Carpenter: I think that’s still a very interesting question, actually. I think it’s still a relevant question. My personal opinion is that OSI as the grand plan has lost its window of opportunity, that we will see components of OSI which will be useful.
Malamud: Which of the components do you have in mind?
Carpenter: Well, I think there is still some chance that X.500 will be useful, the director service, because I think the Internet DNS has limitations. I’ll probably be shot down if anybody hears me saying this, but I think…personally I think one of the options which is viable for the next generation of IP is the TUBA option, which uses components of OSI. I have to say that carefully because that’s not the only option, and it may not be the one that’s selected. And I think there are little components like that, you know. ASN.1, the Abstract Syntax Notation from OSI is an extensive use in the Internet.
Malamud: Are these pieces going to be integrated into the Internet protocol suite or will they be separate? Is this ships in the night?
Carpenter: I think it’s going to be more than ships in the night. I think there’s going to be a degree of integration in the sense that the Internet people have the habit of taking what’s good from the other standardization bodies. I mean they take what’s good for the ITU standardization and what’s good from the IEEE and the ANSI. And they’ll take what’s good from the ISO process. But that’s not to say that they’re taking OSI, that they’re buying the whole bundle. It means their taking the good components. And you know, there are people who have very emotional reactions on this topic and say ISO is bad, you know, and don’t want to hear about it. But I think that’s not going to prevail. I think what’s going to prevail is a pragmatic approach, taking what’s good, and ignoring what’s bad.
Malamud: There’s been a move by the Internet Society to establish formal liaison with ISO and with the ITU, and to do that as a formal liaison in which representatives are sent back and forth and papers are exchanged. Do you think that’s a healthy move?
Carpenter: I think that depends entirely on the deal. It depends on the deal because if the deal is one that constrains the IETF process…that’s to say the way the Internet Engineering Task Force gets to rough consensus and running code, which is the criterion for something being accepted—it’s a very pragmatic approach to standardization. If that process is constrained by liaison with the other organizations then I think it’s a bad thing, because that is the real outstanding quality of the Internet standardization process. If the effect of liaison is actually to feed the other way such that our good ideas get taken up by the official standardization bodies, I don’t see how it can be harmful, so long as we keep change control of what we’re doing.
Malamud: Do you think an ISO stamp on the Internet standards is a useful thing in some countries?
Carpenter: I think to the extent that you can not really expect politicians to fully understand technical issues, I’m afraid it may be that in some countries the rubber stamp would make a difference. It would change the perception of the people who are providing the money, and in the end we need the money to do things. But I don’t think industry’s going to care about that very much. I think it’s only government-funded activities where this is really a consideration.
Malamud: Now the International Organization for Standardization, ISO, is a private body. As is the Internet Society. Could the Internet Society provide that stamp by itself or does it need that ISO liaison?
Carpenter: I think that’s a presentation issue. I think the ISOC is too young, it hasn’t got an image at a political level yet. ISO has got an image at the political level, and the bodies that are members of ISO are things like the national standards organizations of the various countries. So, although ISO is not a government organization, it has members that are government organizations. So I think there is an issue there. I think the ISO rubber stamp is probably still more powerful, politically, than the ISOC rubber stamp; that the ISOC process, the IETF standardization process, is more powerful technically than the ISO process. So that’s a dilemma.
Malamud: For a long time, CERN was the backbone of Europe, the de facto backbone. There were a few other sites; there was University College London, there were some others that were providing connectivity. But over the last couple years, something called the EBONE has come together. How did that happened at the center shifted from CERN over to this broader consortium?
Carpenter: Well, I’m not entirely sure we were ever the only center in Europe. I think that’s quite unfair on some of our colleagues in other countries. For example since we’re in Amsterdam I think it’s fair to say that the Amsterdam people—how to say, the complex of university departments here have been very much in the forefront of networking. And as I said, so have our Italian colleagues. And of course so has the JANET network in the UK. But yeah, we were clearly a major center in terms of where the bandwidth terminated. And this is less true now. And the reason that is, I think that it clearly became obvious to many people that it’s not actually the job of a physics lab to provide general-purpose connectivity. You know, we’re not funded to do that, and we never spent any of our physics money on doing that. We just…offered floor space, essentially, to other people. And it was a sort of evolutionary process. And when people began to think about this as a managerial issue, they realized this was not really the right way to do it.
It’s my job to make sure our users have the best possible connectivity, and one way of doing that is having all the network links in Europe terminate in our computer center, but that is really not the right topology. It’s not well engineered. And it’s not right politically because we’re not there for that, we’re there to do physics.
So this is a very personal opinion. I think it’s a natural development that this general-purpose infrastructure has developed so much in Europe in the last couple of years, such that the general-purpose users are no longer dependent on a disciplinary network.
Malamud: What changed two years ago that would allow something like the EBONE to be born? Or why wasn’t it born four years ago?
Carpenter: Well, as you know of course, these things always happen in Amsterdam because I happen to remember that you were present at the meeting in the Amsterdam zoo—this is true, listeners— in the Amsterdam zoo, which was the formative meeting for the EBONE network. I think what changed actually was that around that time, people got fed up with waiting for the so-called official solution to appear. And EBONE was a consortium of people with common interests. It was not an official proposal. There is an official network now also in Europe, also operational or nearly operational called EuropaNET, and it’s going to be very interesting to see over the next twelve months how much of the traffic migrates from the EBONE infrastructure to the EuropaNET infrastructure.
Malamud: Well let’s compare those two networks. How do the networks differ for example in who their members are? Who joins EBONE, who joins EuropaNET?
Carpenter: Well EBONE is a club, really. And like any club you know, you pay your subscription and then you get the same rights as all the other members. So anybody can join EBONE, as far as I know. I can’t really speak for EBONE, you need to interview Kees Neggers for that. But my understanding is that there is no acceptable use policy which would cause anybody any problems on EBONE. If you’re prepared to pay for your access point, then you can get access.
EuropaNET is sponsored initially by the European Commission. It’s been created in the context of RARE, which is the association of the European research networks, and specifically of the national networks—that’s to say the equivalents in each country of the NSFNET in the States—and it’s offered by a commercial service provider rather than by a consortium of individual sites.
But again, as far as I know anybody can negotiate for a EuropaNET access if they want one. And at some point it will be fully funded by user payments. Although I think there is some startup funding, actually I’m not too clear on the startup funding for EuropaNET.
Malamud: Now they’re both transit networks. EBONE is based on two megabit leased lines. What is EuropaNET based on?
Carpenter: As far as I know it’s based on a hub structure, with a couple of hubs. I’m not the real expert on this so I can’t tell you exactly how many hubs and lines out to the access points from those hubs. Whereas EBONE is actually a ring which constitutes the backbone with lines into that ring from external backbone systems.
Malamud: EuropaNET as I understand it is based on two megabit X.25?
Carpenter: Well EuropaNET is supposed to be migrating to a multi-protocol, with genuine native IP as well as X.25. So I think at the moment it’s in a bit of an intermediate state but it’s supposed to come up later this year, if I have the dates right, with native IP. So that’s…you know, that’s progress in terms of technical flexibility.
Malamud: Can Europe have two backbones? Is that a good thing to have?
Carpenter: Well we’ve got at least two because there are other networks around like some real commercial service providers of IP networking. So…I don’t see why not. I mean, we’re supposed to live in a capitalist society in Western Europe these days so, why not have two? Why not let the user sites choose?
Malamud: There’s a tremendous amount of activity and noise in Europe now about ATM. Do you see that as a major service in the next year or two?
Carpenter: Certainly not in the next year or two. I think it’s going to take a long time to see ATM deployed in a general sense in Europe. What’s happening at the moment is a large number of the operators, I think it’s almost twenty now, signed a memorandum of understanding about the deployment of ATM on a pilot basis across Europe. That’s a very interesting development, but one thing I have learned in dealing with the PTTs is if something is called an experiment…they might just give it to you free. If it’s called a pilot, you’re going to pay for it. Not at full tariff, but you’re going to pay for it. So you know, even at this pilot phase there is a probability that people will have to pay to get international ATM connectivity. It’s not a certainty and it will vary from one country to another.
And I’ve said this before in a talk in London last year, I think the critical issue for ATM is not technology. We’ve got technology coming out of our years. The critical issue is will the price performance of ATM be interesting to the big paying customers. Will the banks, the insurance companies, the airlines, the people try to do remote CAD like the car companies, will all these people want to pay for ATM? Will it be cheap enough to interest them. It’s clear the academic market is not going to support ATM.
Malamud: Many people have said that ATM provides general-purpose connectivity. It provides that cloud into the world that lets you talk to anybody; that doesn’t mean they have to talk back. But it provides that same general-purpose connectivity that the Internet does. Does ATM make the Internet irrelevant?
Carpenter: Absolutely not. I mean, I think the ATM connectivity, if it really appears as the universal connectivity, will be a very interesting way to carry Internet traffic. I think it’s just another technology. I think it’s not the universal solution. I don’t believe I’m going to have an ATM connection to my toaster. I think you know, there’s still going to be subnets, if you want to get technical, which are non-ATM for twenty, thirty, fifty years. I mean, as long as you like. And I don’t believe I’m going to get ATM to North Africa or Eastern Asia that quickly. And I think that’s very important. The Internet is really a global network.
Malamud: You mentioned your toaster. How’re we going to connect your toaster to the Internet? And are we?
Carpenter: Well. I mean, let me tell you a story. A couple of weeks ago there was the storm of the century in the particular area where I live, just localized. And you know, the volunteer firemen were out all night pumping our peoples cellars. And my neighbors happened to be out at a graduation dinner for their son. And they got home at midnight to find two feet of water—about sixty centimeters since we’re in Europe—of water in their cellar. And they were a bit annoyed about this because they were moving out two weeks later and had to hand the house back to the landlord.
Now, it occurs to me that it would have been perhaps quite convenient if their beeper had gone off when they were down at this graduation dinner saying you know, “There’s a problem in your cellar.” And they could have done something about it and maybe got the firemen around a bit sooner and pumped it our before all the rugs were ruined. So I think instrumenting your house is not completely frivolous. If we are in an affluent society you know, this is the sort of thing we might want to do. I’m not too sure about the toaster; that’s…become an Internet joke. But you know, instrumenting my house, yes why not.
We have on the table at least four proposals for the next generation of IP. They will have their technical features. I’m sure they will all work within their own terms of reference. I’m not sure yet that any of them actually solves this problem of instrumenting every cellar in every house in every country. You know, it’s not obvious to me that any of these solutions will really go to that level. Maybe it doesn’t matter. Maybe that’s the wrong approach to solving that problem; I don’t know.
Malamud: There’s two very large sets of activity in the Internet—there’s many others, but one is focusing on the question of the next level of the network layer. There’s another group of people, including Tim Berners-Lee at CERN that are looking at much higher level issues of how to organize information, and find information. Are these two problems related?
Carpenter: Well, in a strict sense, in a mathematical sense, of course they’re not related because they’re occurring at different layers of the OSI model. And the OSI model is useful for thinking about these things. So you can argue that they’re not related. But I’m afraid they are related because of course if you start building things like the World Wide Web, this presumes that the network is fully connected. And when I find a pointer in a hypertext document that points to another document that is in Australia, the World Wide Web model presumes that I can go look at that document and get it in a second or two of real time. And that’s a very big presumption about the underlying infrastructure. Because we’re all used to the Internet being partially broken. I mean, every day the Internet is partially broken. There are connectivity holes for one reason or another. And if you’re putting higher-level applications on the Internet, and these are very important applications, that assume the network is not broken, then you’re going to have very frustrated users—much more than today.
If I do telnet to a machine in Australia and it doesn’t work, then I try again later. If I’m in the middle of a World Wide Web session and I happen to click on a button that follows a pointer to Australia and it doesn’t work and I get very annoyed and I keep clicking on that button until the mouse is worn out.
Malamud: So we need to be thinking in terms of robustness and the production quality of the Internet.
Carpenter: I think that— That’s right. I think this new generation of applications, and it goes for any distributed computing application, assume a robustness of the underlying layers that is not necessarily the case today.
Malamud: So should the PTTs then take over those underlying layers?
Carpenter: Uh, am I allowed to say “no comment” in this interview?
Malamud: You sure are. Thank you very much, Brian. This has been Brian Carpenter. He’s at CERN. He’s the group leader for the Communication Systems Group which runs the CERN network. This has been Geek of the Week.
This is Internet Talk Radio, flame of the Internet. You’ve been listening to Geek of the Week. You may copy this program to any medium, and change the encoding, but may not alter the data or sell the contents. To purchase an audio cassette of this program, send mail to firstname.lastname@example.org.
Support for Geek of the Week comes from Sun Microsystems. Sun, the network is the computer. Support for Geek of the Week also comes from O’Reilly & Associates, publishers of the Global Network Navigator, your online hypertext magazine. For more information, send mail to email@example.com. Network connectivity for the Internet Multicasting Service is provided by MFS DataNet and by UUNET Technologies.
Executive Producer for Geek of the Week is Martin Lucas. Production Manager is James Roland. Rick Dunbar and Curtis Generous are the sysadmins. This is Carl Malamud for the Internet Multicasting Service, town crier to the global village.