Carl Malamud: Internet Talk Radio, flame of the Internet. This is Geek of the Week and we’re talking to Bob Braden, who’s at the Information Sciences Institute at the University of Southern California. Bob, welcome to Geek of the week.
Bob Braden: Thank you.
Malamud: You’re the director of DARTNet, which is ARPA’s…experimental network out there. Could you maybe tell us what DARTNet is and…what is it? I mean how does that relate to the rest of the Internet?
Braden: Okay, well director’s maybe a little strong.
Braden: Coordinator is more accurate. I sort of fell into that because the DARTNet work came out of discussions in the End-to-End Research Group of the IRTF that I chair. I’ve chaired it since the start.
Malamud: That’s the Internet Research Task Force.
Braden: Internet Research Task Force, right. And it is was a hotbed of Internet-related research pepole. And these people originally had the ARPANET to play with. And then they had the Internet play with. But the Internet became too successful to be a research vehicle anymore. And people don’t like us to break it. So ARPA agreed to create a network which we could break. And that’s DARTNet. It’s a cross-country TI network, a very simple backbone spine, that connects some ten research sites. We use Sun SPARCstations as packet switches. A very important part of it is that the packet switches are programmable. And so to do an experiment on DARTNet you typically move your own version of the operating system, with its own network code, and take over the whole system—the whole network—and do experiments.
Malamud: And what kind of experiments are done on there.
Braden: Well, mostly experiments where you need to break the network. The center of interest has been in experiments concerning the dynamics of networks. Issues of congestion, of multiplexing, of queuing, scheduling, now of resource preservation particularly. There have been some experiments but not a lot in routing, because it doesn’t have a very interesting topology and so it isn’t terribly useful for routing. And in factof the kinds of things you want to do in routing besides simple demonstrations that it works involve scaling issues, and for that you really need stimulations because you can’t afford to put together a hundred-thousand-node network for scaling experiments. But you can do a lot of really useful scientific work on packed scheduling kind of issue. If you get a bunch of packets of different traffic classes, which ones get forwarded. So general generally that’s been the area.
Malamud: So that’s the area of resource reservation or…is this really are we reserving bandwidth here, or are there are other techniques that are being looked at?
Braden: Well, there were experiments with fair queuing, which is not reservation but has to do with the dynamics of traffic, and congestion. And more recently we’ve been doing explicitly resource preservation experiments. Our goal is to develop a technology which can be used in the entire Internet to allow us to do voice and video acceptably, with reasonable quality across the entire Internet.
Malamud: There’s a conventional wisdom that says the Internet has worked and it’s fine for non-real-time services: mail, and FTP—
Malamud: —and one can even consider telnet to be somewhat non-real-time. But when it comes to voice and video, that won’t work and we’re gonna have to move back to a connection-oriented system. Do you think the Internet is going to have to make that about-face or we’re gonna have to kind of get rid of packet switching and move back to connection-oriented networks?
Braden: Well, don’t think anyone believes that anymore. There has been— Well. You need to distinguish the discipline you use for forwarding packets—who goes in what queue and when you fill and when you empty the queues—from the question of the setup protocol—the way the way you actually make the reservations. For the first issue, the scheduling of packet forwarding really doesn’t…isn’t connectionless…can be done a connectionless way. We can do this with IP packets, IPng packets, or CLNP packets. It doesn’t make any difference.
But you need state in the switches which reflect the classes or the sessions for which you’re gonna reserve bandwidth. And there’ve been two general approaches suggested for setting up that state. One is a connection-oriented approach represented by SD2[?]. And now we’re currently developing a connectionless approach called RSVP, and that’s the work being done on DARTNet now.
And over the years we’ve tended to come to believe very strongly in connectionless approaches as providing more robust service, and service which allows plug-and-play, which allows people to connect stuff together, built by a variety of different vendors with only a modest amount of careful engineering and have it work correctly.
Malamud: Well how does RSVP work? What are some of the techniques you’re looking at to do this connectionless resource reservation?
Braden: Well basically you send packets along the path of your data to set up the reservation, and that creates state in the routers along the path. And that state times out. There’s no hard mechanism to remove that state; it simply times out. And you refresh it. You periodically resend the same data before it times out. And if for example a route changes, then the refresh packets follow the new path and set up the state on that new path. And the state on the old path which is now not in use times out. So it’s all done on a sort of soft-state basis.
Malamud: And do these packets guarantee that you’ll have a certain amount of bandwidth? Is it a best-effort guarantee, if one can use that oxymoron?
Braden: That’s an issue of the scheduling algorithm you use, the model for traffic control. And that’s independent of the way you do the actual setup. The setup article carries along what we call flow spec, which defines the quality of service you want. And you may of course get back an error message saying, “I’m sorry, I can’t provide that service.” But it’s still pretty much an area of research what the qualities of service need to be. There is a model which is very popular which is developed by Dave Clark at MIT and Lixia Zhang and Scott Shenker at Xerox PARC, we call it the CSZ model, which is the one we’re pursuing but there are other proposals.
Malamud: If we’re looking at classes of service and reserving resources, or certainly at least saying we’d like some resources, the Internet currently is based on flat fees. Users don’t pay, or users pay a certain amount per month. Does that mean if we have different classes of service every single user is going to say, “We want the best possible service of course?” And how do we handle that with the current charging model? Is there a dichotomy there?
Braden: Well that’s a very important issue. The researchers in this area believe, quite clearly, that resource reservation can only work if there is some feedback to the user. Now, that doesn’t have to mean charging. But there has to be some felt cost to the user with the ask for a better-quality service. Because otherwise as you say, the system would just break down; everyone’ll ask for the best quality of service and then nobody’ll get anything. So that some sort of a usage-based resource feedback is essential. And well, we say that and so far we haven’t done anything about it.
Malamud: What kind of feedback might that be? obviously taking your money is a good way of doing feedback. Are there other ways of doing that? Does the screen get muddier if you want better service?
Braden: Well of course it’s probably true that the Internet is gonna move towards a real-dollar basis for most people. So at some level there’s going to be money exchanged. It may be… Well. When I go to work and I pick up the phone to make a phone call, I’m not aware directly of the cost of that call. Now, I may get a monthly summary from my company saying that my department spent so much or even my telephone spent so much. If I start making a lot of long-distance calls after hours to Japan, or if the phone bill begins to mount up in some particular area, some administrator may ask you know, what’s going on. The company may decide to buy a different local phone system. A different PBX which is able to optimize the traffic in some way, like it tries different services or gets the best—the cheapest service, is able to compute which carrier to use at what time of day.
I think that network charges will probably be analogous to this. That companies will get bulk charges which they may or may not reflect—probably generally reflect down to the individual user, but there will be someone in charge who is looking at the costs and optimizing them. And deciding that well gee, maybe we ought to buy this better FTP program which which will save us 20% on our average costs. Or maybe we can’t afford quite the same quality of service as a whole. So there will be some indirect administrate feedback. I think most people won’t see it directly.
Malamud: If we’re looking at controlling costs in a network, or assigning costs, we’re implying accounting in some respect. How are we going to account for the use of resources in the Internet? Is this going to be something that every single router will maintain a log of every packet. Is it going to be simply the endpoints where we keep a log of FTP sessions? Or is it someplace in between or someplace different?
Braden: That’s a controversial issue. And we haven’t…really thought enough about it, or done enough experiments. So I don’t really know the answer except that we believe in general…what it costs in some sense, by some metric whether it’s real money or some usage measure. What it costs has to reflect what you do, what you ask for. If you ask for a better quality of service, it has to cost in some sense more.
Malamud: And any ideas on how we might do accounting? Because right now the Internet is based on a—this is probably unfair, but a no security, no accounting, open model. How might we begin doing accounting? Do you have any at least ideas if not answers?
Braden: I think a reasonable model might be the sort of usage feedback is only necessary for people who ask for better-than-minimum service. So if you only ask for the vanilla best-efforts service, we may not bother. But if you ask to reserve bandwidth for a video signal, now that’s a significant resource. And it needs to be accounted to you in some way.
As I say, we— Well. The fact that you have to ask for reservations, whether you do it in a connectionless or a connection-oriented way, in either case you have to explicitly set up a reservation. And that act presumably triggers some sort of accounting. Now, whether you actually count packets or whether you only— Whether the cost is based upon usage or upon reservation is one of the basic question. And there are arguments both ways. In other words if you reserve 64 kilobits over certain path, and you have the right to use 64 kilobits, do you get charged on that basis regardless of whether you use it, or do you only get charged on the packets you use?
Malamud: You’re looking at a variety of research issues on how the network will function. We’re currently going through a process of engineering the next generation of an Internet protocol.
Malamud: And that’s based on a crisis in the address space, and in the routing tables. There is some disagreement as to whether we should just solve that address problem, or whether we should at the same time while we’re going through that transition try to tackle some of these more long-term issues. How do you see that coming out?
Braden: Well I think that the address-based problem. which is actually real, is getting a lot of attention because people understand it. The network operators understand it very painfully and they see that coming. Or least the limitation of their growth, and for them that’s a very serious issue. But I think that people are not paying enough attention to the explosion of that, the use of video and audio, and the fact that all our workstations now are beginning to come with audio and video built in. And soon tha’ll be absolutely standard. And that’s a tremendous traffic generator and this stuff is nice—people use it. We use it because we’re developing this technology in DARTNet. A number of us on DARTNet have an open channel for audio at all times when DARTNet’s not being broken by somebody. And…we just use it. And we use video quite a lot and it’s very helpful.
s. So by an open channel, it’s as if you’re in a room with all these other people—
Braden: Essentially. Except we keep our mics muted most of time. And if I want to talk to Steve Kasner[sp?] on the other side of my building or Steve Deering in Palo Alto I just say— I can’t just say “Steve,” I have to say, “Steve Deering are you there?” and we chat. And that’s very handy.
Malamud: And that’s going to be…all over the world, that same type of service.
Braden: I believe so. And now, travel is just horrendous. We can’t do all— There are too many meetings and too much travel. And the only thing which is standing in the way at this point is the quality is rather poor because we don’t have bandwidth reservation. As soon as we begin to have bandwidth reserva— Well the other thing is that people are more and more beginning to use this on their LANs. It going to become more and more popular on LANs. And then they’re going to say, “Well why can’t we use this the wide Internet?” So I think that very shortly there’s gonna be great pressure on the vendors and on the operators to provide resource reservation service Internet-wide.
Malamud: Well how are we gonna do that? Do we know how to do that?
Braden: Well, that’s the research which I’ve been talking about on DARTNet. And our intent is to provide a set of algorithms and prototype routines and protocols which then can be introduced into the IETF process. And I believe that— My current reading is that it would be a very good idea to get that in sync with the IPng work.
Malamud: There’s a sense of urgency about IP Next Generation, about getting it out there and avoiding the address crisis. Will the resource reservation work be ready in time or would that slow down the process of deciding on the next IP? Can we afford to wait for resource reservation?
Braden: Well, that’s… Would you like to peer into my crystal ball. As noted at last night’s IP [indistinct], there’s a great deal of uncertainty about how soon we have to we get IPng developed. It looks like CIDR’s going to save us for a while, but we don’t know how long a while is. I guess I— It seems clear— We need to move ahead with all deliberate speed on IPng. And I think we have a chance. I think we have a chance to get the resource reservation stuff far enough along in time that it can be factored into the decision and development of IPng. Our goal has been to begin to introduce the resource reservation work into IETF— Well actually, at one point our goal was to do it this fall. We’re not gonna make that, but by early next year; early calendar ’94. And I think we oughta synchronize the two efforts.
Malamud: How long do we have before the addressing and routing table explosion problem…kill us? How long do— How bad is the crisis? I know you don’t have an answer but do you have a range? Are we gonna run out of addresses in fifty years? Are we’re gonna run out next month?
Braden: Well I think the numbers which have been mentioned recently range from what, two to five, six years. And [indistinct; crosstalk]—
Malamud: That’s if we deploy CIDR and if it works.
Braden: Yeah, assuming CIDR, yeah.
Oh, CIDR will work, but how effective it will be we don’t know yet.
Malamud: You’re executive director of the Internet Architecture Board, and in fact you’re a long-time member of the IAB.
Braden: A charter member, actually.
Malamud: Charter member. That’s about as long as you can get.
Braden: Since 1981 when it started.
Malamud: How does a group like the IAB monitor the tremendous output coming out of the IETF? How do you judge and review security and routing and resource discovery, and X.500 at the same time?
Braden: It was only in the old world order that we did all that. And the answer would have to be fairly poorly. But in the new world order, we play much more of a management kind of a role. We are tasked to worry about—or to do some long-range planning, try to step back and think about the architecture and about the general principles of what’s going on. And we also are tasked to serve as a sort of court of appeals, basically, to that the IESG that has primary responsibility on standards.
Malamud: Do you see that function being…[crosstalk] exercised a lot?
Braden: So we watch an listen a lot.
Malamud: You watch and listen, and if necessary use you step in when the IESG can’t make a decision or was challenged its decision.
Braden: Well that’s the principle. I mean in fact it hasn’t yet happened, so we don’t know how well that’s going to work. But that selection process I think is really important. And it’s really important to the community that that work.
Malamud: The Internet has always been a real dynamic laboratory, a place where new things happen and we’re not afraid to try to move forward. And now it’s becoming a global production service, very operational. We’re seeing telephone companies offering IP service. Is that gonna change the Internet? Is it gonna be a different place?
Braden: Well that’s an imponderable. When there’re such big bucks involved it’s hard to see how we can be allowed to continue to play the way we have.
Malamud: Isn’t playing good?
Braden: Well I mean, we’ve been tremendously successful. We’ve built a great thing. Without wearing suits and ties. But there’re an awful lot of suits and ties in the world, and when there’s such big bucks involved and big political— When the Vice President of the Unites States starts talking about it, and when you find jokes in The New Yorker which talk about the Internet with no further explanation, you know you’re in trouble.
Malamud: This is dog joke that—
Braden: Yeah the dog joke, right.
Malamud: There was a cartoon in The New Yorker in which one dog is sitting on a chair and looking at a terminal, and there’s another dog on the floor, and the dog on the chair says, “On the Internet, nobody knows you’re a dog.”
Braden: [groans] But what was remarkable about that was it appeared in The New Yorker and it used the term “Internet” without any comment, without explanation. And it— Well, I read a lot of articles about the Internet today in The New York Times. It’s part of the discourse. It’s becoming part of discourse of the nation.
Malamud: You started working on the Internet when?
Malamud: Were you part of the original ARPANET team?
Braden: Yes. Well I actually started networking back in 1970 when the ARPANET started. I was at UCLA and in charge of systems programming for what was a supercomputer in those days, an IBM 360⁄91. And ARPA wanted to have a supercomputer on the ARPANET as a resource, because ARPANET was supposed to be about resource sharing; remote access to computing resources. Which were then very expensive, relatively. And so they came to the computing center at UCLA and asked us if we would make our IBM system a host on the ARPANET. So, I was in charge of that and I got interested in the problem of protocols and got interested in— I worked on FTP protocol and was responsible for implementing the whole suite. I didn’t do the work at that point. I was a manager. But I went to working group meetings and got to know John Postel and Steve Crocker and the very creative group of people who started all that. So I have some number of RFC written back in the 100s, the 200s in the early 70s. And we were something like fifth or I don’t know, tenth host on the ARPANET. And about 1973 I guess we are fully operational. Those were very exciting days, interesting days.
Malamud: Did you have any idea that the Internet would grow like it did?
Braden: Well. This was the ARPANET, and the ARPANET grew in sort of predictable ways. But then in 1974, the Internet idea was developed at ARPA by Bob Kahn, and Vint Cerf wrote the classic paper on Internet protocols in 1974, and then 1975 they started the Internet research program. And about 1977 I guess, I got an ARPA contract to change our host software at UCLA to support TCP/IP. So I wrote the TCP/IP for the IBM system. And this time I did type the code.
And so I became part of Internet research program at that point. I was actually in the TCP end. Vint went to ARPA. He separated the IP and the TCP groups to work separately and I was in the TCP group.
Malamud: 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 email@example.com.
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 firstname.lastname@example.org. 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.