Intel Developer Forum, Fall 2004

Pat Gelsinger
San Francisco, Calif.
September 9, 2004

ANNOUNCER: Ladies and gentlemen, please welcome back Frank Spindler.

(Applause.)

FRANK SPINDLER: Thank you very much. Welcome to day three of IDF.

I'd like to thank all of the sponsors, exhibitors, presenters, staff, attendees who have made this such a great week so far. And in particular, I wanted to acknowledge and thank the hundreds of you who have come from over 40 countries around the world to participate in what we think has been a great week.

But we're a long way from being done. We've got a full day ahead of us today. We're going to lead off with Pat Gelsinger, who's going to talk about the new Net, give some intriguing thoughts about how the Internet will evolve and what sort of next-generation capabilities that can bring us.

We once again have the technology showcase open for a couple hours, from 11:30 to 1:30, and a full range of Birds of a Feather sessions at lunches today to share in and share ideas with industry experts on a variety of specific topics. And then more sessions and hands-on labs as well.

I also wanted to thank everybody for their patience yesterday when we had a brief power outage in some parts of the building. So we appreciate that.

Now another thing we'd like to ask, is that you mark your calendars. IDF spring will be here before we know it, February 28th to March 2nd back here in San Francisco, back here in Moscone Center. So set those dates aside, please.

I'd also like to ask you, in the spirit of helping us to continually enhance and improve IDF, is the attendees will be receiving E-mail surveys asking for your feedback on the conference. And so we would very much appreciate if you would fill those out and send them in to us. Because it really helps us direct the future of IDF.

So without further ado, let's get the day rolling. And I would like to introduce our Chief Technology Officer, Pat Gelsinger.

(Applause.)

PAT GELSINGER: The year was 1973. "Bad, Bad, LeRoy Brown" was the number one hit song of the year; "The Godfather" won the Oscar for best picture; Secretariat won the Triple Crown. And I was in sixth grade.

(Applause.)

(Laughter.)

PAT GELSINGER: But while those events were going on, something much more important was happening. At my alma mater down the street at Stanford University, two gentlemen, Bob Khan and Vint Cerf, were laying the foundations for what became the Internet.

Ladies and gentlemen, Vint Cerf.

(Applause.)

VINTON CERF: Pat, a pleasure to see you.

PAT GELSINGER: Great to see you, Vint. Thank you for joining us, Vint.

Vint, 1973. All right, lots of things going on. And you were holed up in a lab there. What were you thinking about? What were you and Bob trying to accomplish?

VINTON CERF: Well, Bob Khan was serving at DARPA, the Defense Advanced Research Projects Agency, in 1973. He had worked on the ARPANET and come to DARPA to pursue packet switching technology as a research topic. And the Defense Department was very interested in using computers in command and control, so they needed to have mobile communications with radio links on ground mobile, and they also needed communication for ship to ship and ship to shore on satellites. So they were trying to find ways of using packet switching technology in these different media. And Bob came to me at Stanford and said, "Well, I've got this problem to try to interconnect these networks." And about the same time period, Bob Metcalfe and Dave Boggs were inventing Ethernet at Xerox PARC about a mile and a half from my lab. So the goal was to bridge together these heterogeneous packet-switching networks to make it possible for computers to send across them.

To send information across this patchwork of networks, the applications and services would have had to know in some detail exactly how the networks were connected, what their topology was, what routes the packets should take, all the details of the physical transmission technology, the kind of delays that might be experienced, and things like that.

The applications and the services would have had to specific all the various protocol conversions to go from one net to another, the router ports and so on.

So we wanted to make it so they wouldn't have to worry about all the details, the underlying details of transmission. So we created TCP, transmission control protocol, and then later we added Internet protocol, which hid all those details from the applications. So with TCP/IP, the applications just had to say, "Here, send this packet to that global destination." So TCP/IP became a suite of higher-level protocols overlaid on top of the existing packet network protocols.

PAT GELSINGER: So your goal was establishing this view, this image, of end-to-end transparency that the edge computers, the application services, all the people that would use that network wouldn't need to trouble themselves by understanding or knowing anything about the underlying physical network; they just sort of drop information into one end of the Net, and, poof, it would voila, appear somehow on the other side of the Net.

VINTON CERF: Exactly. Somehow like dropping a postcard in the post office, you don't know what happens in between, but you drop a card and it gets out on the other end.

PAT GELSINGER: And you did that without replacing all of those underlying networks by putting something on top of it.

VINTON CERF: That's right. We just superimposed TCP/IP on top of all the existing networks and caused them to be interconnected using what we call gateways at the time. Now people call them routers. But that allowed us to implement a powerful new level of functionality without having to rebuild all the existing networks.

PAT GELSINGER: Very interesting. So who was going to use those networks and what were they going to use it for?

VINTON CERF: Well, the original ARPANET goal was to supply access to computing and communications software and access to other people's programs through this wide area packet-switched net.

So DARPA was funding computer science research at a number of different institutions around the United States and build the ARPANET to link them together so they could share resources.

The initial goal of the TCP/IP, on the other hand, the Internet, was to allow military to use computers in command and control from tactical environments all the way back to strategic environments with computing in the continental United States sitting on the ARPANET, for example.

So in these early days, the applications that we had processor pretty limited. We had FTP, TELNET, and E-mail.

PAT GELSINGER: Wow. That's pretty amazing. So, you know, we had, you know, some research projects, some fairly modest Defense Department goals, and those -- that initial work has just grown and flourished to all sorts of new users and new usages. It's become this vast public and private network that spans the globe, right, and we're approaching, you know, a billion users on that network daily. And not just the usage, but the vast amount of different applications and services that have been built on top of it, you know, today audio, telephony, video conferencing, et cetera.

VINTON CERF: Yeah, it is pretty cool, isn't it? Especially, when you think about it, this architecture is now over 30 years old.

The basic protocols are still pretty much unchanged from what we specified back in the 1970s. In fact, the IP version 4 that we're generally running today was standardized in 1978, 26 years ago.

New protocols, of course, have been layered on top of TCP, IP, and UDP in order to support new applications.

PAT GELSINGER: I just find that stunning, that most technologies maybe have life spans of five or ten years. I mean, my chips, right, we did 386 and replaced it with the 486, and then the 486 gets replaced. My greatest inventions haven't lasted 30 years.

So I just think it's amazing what you've been able to accomplish.

So what do you think of the Net today?

VINTON CERF: Well, to be really honest, I think it's still pretty primitive. I think that we're still kind of in the Stone Age when it comes to serious networking.

PAT GELSINGER: Now, I find that pretty amazing. And I think a lot of people would be shocked to hear the inventor of the Net say it's Stone Age, right, considering all that it's accomplished and, you know, adapted to and the way it's been able to do that without breaking.

VINTON CERF: Well, maybe I should have said "Silicon Age." But in any case, you get the point.

Look, there's a great deal for this has to be done. We have to explore alternatives and variations on the existing architecture. And we need a way to do that?

PAT GELSINGER: So you think the limitations are architectural?

VINTON CERF: I think so. I mean, there are a lot of different kinds of limitations on the net. But some of the more fundamental ones I consider to be truly architectural.

PAT GELSINGER: Mm-hmm.

So, you know, I'd like to look a little bit, Vint, forward a bit at what some of the new, you know, drivers and trends that will change and force that next Net into consideration. We're looking at the sheer number of users. We have a range of devices, and we continue -- and IDF, if there's anything that IDF does, it's cool gadgets; right? We're innovating all sorts of new things and new tools for that, in fact, I pulled one out of our lab before come down here. And this is -- what we did here with this is, you know, this is a full Windows-compatible Intel machine that's less than a pound, right, full keyboard capability. This --

VINTON CERF: That's weight, not price? Too bad.

PAT GELSINGER: And what we're trying to do with this is explore taking the platform to more places. Imagine a doctor in India, right, for cost concerns, networking concerns, et cetera, but wants to deliver the world's best health care services, right, and be able to do so with the information that's available on the Net.

So we continue to innovate around these new devices. And these are some of the things that we want to bring to the network in the future.

VINTON CERF: Of course, there are all kinds of implications of having devices like this so conveniently available.

The Net is going to need to accommodate a lot more people, a lot more terminations on the network. Let's face it, the vast majority of the world's population today still doesn't have access to the Net, and ideally, we'd like to bring those billions of people currently excluded from the Net online.

PAT GELSINGER: Yeah. We at Intel call that the next 5 billion.

VINTON CERF: Well, we need to connect the next 5 billion, you're right. And we also need to connect a lot more devices on the network, things like appliances, sensors and even cars. And at the same time the Net has to support many more points through which connectivity can occur.

For example, wireless hotspots are a way that people can have access to the Net or maybe mobile phones with GSM or CDMA or network-enabled devices in general.

PAT GELSINGER: Then, of course, we want a whole class of new applications that gets supported by those networks and devices.

VINTON CERF: This is what really excites me. There are a huge number of existing services that are in the physical world that are still beginning to migrate into this virtual network environment. You mentioned already health-care services, for example, or entertainment and gaming, financial transactions, security and environmental control. All kinds of new and demanding applications that are designed for the Net but require a lot more capacity. Things like high-quality video conferencing or HDTV broadcasts or massive multiplayer games.

PAT GELSINGER: In fact, one of the technologies that we've talked a lot about here at IDF, and we had a bouncing ball demo a couple of years ago, was sensor networks. This idea that we wanted to create a whole new class of devices, and that those would connect up to the Net somehow. And those might be RFIDs, passive devices or active devices. And I have one here of our sensor network nodes that we showed up, and it's communications, it's a complete radio so it has a whole RF stack; it's storage so it can have big programs; and it's a computer built into it; all battery operated, small, and you stick it in all these different places and we've had some fun demonstrating them here at IDF but we've worked at putting them into real-world applications.

And this is really exciting as we started to see those applications emerge, and in fact the picture that's being shown here in the slide is an Intel fab where we literally started to populate these into our fab to gather environmental data that you referenced before to help monitor and control our fabrication environments. And it's getting great acceptance, and data that we could have never gathered before.

We're also working with others in the industry and we're very happy to have a partnership, a research partnership that's now going to real-world deployments with British Petroleum. And if we could take a look at the work that we're doing with British Petroleum right now.

VINTON CERF: I'd like to see that.

(Video playing.)

PAT GELSINGER: BP putting sensor networks, real-world applications, vast quantities of data flying at the Internet. And as you heard, significant economic benefit.

We're very pleased with the partnership with BP, and in fact we're delighted to have Lord Browne who is the group general chairman of BP joining us in the audience today. And featured in a number of magazines recently for the great success of BP. And not only is he the group general chairman he's also a member of the Intel Board of Directors. And we're delighted to have you representing BP here with us today. Thank you, Lord Browne.

(Applause.)

VINTON CERF: So with all due respect, Lord Browne, I have to report a much more important application than detecting the amount of fuel in the tanks. And actually, this is one that you don't know about, Pat.

I have a wine cellar at home, and the humidity and temperature of that wine cellar is a vital concern to me. And it is my intention to use the motes to attach the humidity and temperature sensors to report every five minutes or so what the state of the wine cellar is, put it up on a website so anywhere in the world I can check to make sure my wine is okay. And if there's a temperature increase beyond a certain rate, my pager is going to go off.

PAT GELSINGER: I got it, I got it.

VINTON CERF: The other thing that is exciting is we've been working at the jet propulsion laboratory with your team at Intel Labs at Berkeley using the motes for the interplanetary Internet design.

We're intending to use this technology or something like it in a radiation-hardened environment to put sensors down on the surface of Mars, run a set of protocols within the sensor mote system, and then attach that to the interplanetary Internet whose protocols we've pretty much standardized. And we're ready to launch in 2009 an orbiter around Mars. So this is exciting stuff.

PAT GELSINGER: Let me present with you your first wine cellar mote.

VINTON CERF: Fabulous. There it is!

(Applause.)

PAT GELSINGER: Vint, all of this is going to have just a profound impact.

VINTON CERF: Well, I hope so.

The one thing that's pretty, well, mind -- thought provoking, I think, is there is going to be a very big socioeconomic and geopolitical change as a consequence of all of this technology flowing into our daily environment.

PAT GELSINGER: And I've been extraordinarily impressed by your ability to look forward in the future, but I think we're not going to be able to predict all the impacts that this is going to have.

VINTON CERF: You and I are absolutely right. It's really all going to surprise all of us, I think.

But one thing that is predictable is we're going to become more and more dependent on the Internet.

PAT GELSINGER: And in fact, it's stunning with the level of dependency that we already have on the Internet. If we just look directly, some surveys have shown that the amount of e-Commerce represents as much as 6 percent of the GDP of the U.S., and if you look at the dependent businesses to the Net for their operation, it's by some reports as high as 22 percent.

VINTON CERF: Well, so what that means is that all of us who are involved in either operating the Net or making components of it, writing software, running on the Intel processors and things like that are going to be responsible for supporting more mission-critical applications, things that are of life-and-death importance. You mentioned the medical health care, for example. Everything from keeping cars and highways running, to maintaining our homes to dealing with vital medical needs.

PAT GELSINGER: And, you know, it's just incredible that when you think that 15 years ago we didn't even know what the Internet or Worldwide Web was I remember when I was working on the P6, what became the Pentium Pro and what was the precursor to Pentium II, III, and 4, one person came over and showed me Mosaic. And when he got done I said, "Do it again." I was so fascinated by what this might mean, reaching out and touching -- I think it was Purdue at the time. And from that first Mosaic, right now a household word that a three-year-old learns --

VINTON CERF: It's pretty stunning. I've heard about eight-year-olds teaching their classes about how to make web pages. Don't look back; there's a 13-year-old gaining on you.

(Laughter.)

VINTON CERF: It's really been an amazing journey, but I think we're still in the very early stages of it.

PAT GELSINGER: Can we talk maybe for a second about some of the challenges ahead and if we're going to realize and support all of these users and usages of the Net, you know, we have some pretty big obstacles to overcome, Vint.

VINTON CERF: Well, I think that's right. In fact, there is a kind of capacity problem. We hear about oversupply of fiber and everything else, but seriously, how is the Net going to support all these billions of additional users and devices and those new data-hungry applications?

The number of devices on the Net will certainly grow into the tens of billions. It's going to exceed the address space of the IPv4 protocol, which is why we need to go to IPv6. The users and applications are going to demand much more overall bandwidth than the current infrastructure delivers.

Beyond bandwidth and more users, you get a potential for larger localized traffic jams. Bigger crowds flock to particular sites. In Net speak, we sometimes call these flash crowds.

PAT GELSINGER: Mm-hmm.

VINTON CERF: And as the total population on the Net goes up, we'll see larger flash crowds concentrating at certain sites. And there's yet a greater problem, greater variations in the number of hops that it takes to get information to different users in different geographical locations. You might be lucky and you're close to a source but somebody else may be far away and the Net has to hop through to deliver it.

So there are going to be widening disparities in transmission time and quality of service, depending on where you happen to be in the system.

PAT GELSINGER: And it's not just capacity but reliability. As the Net expands you just have more points of failure. Performance becomes more inconsistent as you look across that network, and as we become more reliant on it, the Net itself is the mission-critical resource of the economy as a whole. And a full fifth of the economy running on it is just incredible. And things like Sobig and MyDoom, huge economic impact by a simple little virus. And you can imagine how catastrophic, five or ten years from now, a network outage or a significant disruption in service would be.

VINTON CERF: The potential is very significant. As we get more users and more critical applications on the Net, we have a larger pool of potential attackers and more points to attack.

PAT GELSINGER: Mm-hmm.

VINTON CERF: So the would-be attackers are getting bolder now because they have access to tools making it easier to launch, for instance, worm attacks, with large numbers of machines that have been compromised.

Then with all the new devices connected on the Net there are new types of security threats, like the cell phone viruses that we're hearing about or Wi-Fi entry and things of that sort. So this security problem is really snowballing. The cert at Carnegie Mellon reported that the number of malicious attacks on the Net has been doubling for the last several years.

PAT GELSINGER: And it's not just those problems but accessibility. Remember, I want to get to the next 5 billion, and some of those are going to be in increasingly remote locations. Those locations may have problems of connectivity. It may be the bus that comes by once a day to connect them to the Net, or they may have unreliable power or other disturbances, and how do we reach and deliver service to that full breadth of user as well.

VINTON CERF: That is the problem. I mean, how do we get to those people and by what means? Is it fiber? Is it radio? Is it satellite?

This explosion of devices and new users brings up yet another problem, and that's a regulatory problem, as governments wade in and try to come to terms with this evolving Internet, especially as traditional telecom services migrate onto the Internet, such as voice communication.

PAT GELSINGER: And, you know, I almost think of it as a freight train of regulatory concerns coming at the Internet. And it's -- you know, lots of these are very sensitive issues that need to be addressed. Right, governments need to weigh in considerations of personal privacy, security, national interest as well, and they'll have to confront issues like economics and taxation as well.

VINTON CERF: It seems to me that it's just, generally speaking, a lot of those issues can be addressed in the private sector.

But some of these problems, like taxation or crime on the Net, really have to be addressed in a governmental framework or an intergovernmental framework between -- you know, agreements between countries to do extradition and other sorts of things.

So here we are, the Net's done a good job. But we have a lot of work ahead of us. And we can't necessarily do it with the Net itself, because we can't go in and modify what's already there and transform the entire network overnight, even the deployment of IPv6 is going to take a long, long time?

PAT GELSINGER: And as we look at these problems, capacities, reliability, security, accessibility, and regulatory, they're different than we've confronted before, and the size and scope of them is almost certainly going to require some level of revamping of the network itself.

VINTON CERF: Well, that's precisely the problem that we face now.

We added faster media on the Net, we put in faster processors, many of them made by Intel. We took the basic paradigm and made it go faster. We put more memory in place and things of that kind.

But at some point, the problems of reliability and security may not be addressable just by making changes to routers and putting in more fiber?

PAT GELSINGER: So, Vint, I have an important question for you. So we need to look at these architectural limitations. So, you know, should I give John Hennessy a call, see if we can reopen the lab at Stanford and have you go back and do it all over again?

VINTON CERF: Well, I wish we could do that. But I think it's a luxury that we can't afford.

The Net's really a big collective entity. It's a federation of many different companies and organizations. And it makes it impossible to just rip everything out and replace the existing infrastructure in a short period of time.

So this kind of evolution of the established infrastructure is going to happen eventually over time, but maybe not quickly enough to allow the improvements we're going to need in the near term?

PAT GELSINGER: So is the situation, then, fairly similar to when we created the initial Internet, where we have all of these preexisting Internet, you know, networks that we couldn't replace? Right? And then, you know, we can't rip those out. So instead of ripping them out, you sort of layered a protocol, TCP/IP, on top of them.

VINTON CERF: Exactly what we did.

PAT GELSINGER: And that's exactly what I want to talk about in the rest of the keynote today, is looking at that next layer on the Internet.

VINTON CERF: That's great. Because you'll take this gigantic global system and use it as a scaffolding to build the next generation.

PAT GELSINGER: Precisely.

Vint, you know, I so appreciate you joining us here at IDF. And, you know, I call this the greatest Geekfest on earth. And from one geek to maybe the greatest of geeks, right, the head of the Internet, my sincerest gratitude for having you join us today at IDF?

VINTON CERF: Pat, thanks for letting me be here. Thanks to all you for what you're doing.

(Applause.)

VINTON CERF: See you on the Net.

(Applause.)

PAT GELSINGER: You know, I just can't say enough about my admiration for Vint Cerf. You know, he's a gentleman, a scholar, thoughtful individual, you know, and just someone of such profound impact on what we do on a daily basis. And what an honor to have him join us at IDF.

And again, thank you to Vint Cerf.

So as we've looked at this, we see that we're at this crossroads, right, this point of inflection as we look at the evolution of the Internet.

You know, we've got more mileage out of the Internet than any of us would have fathomed when we started it. But as Vint said, we're running up on some architectural limitations.

Here's a quote that I think sums up our current situation very nicely.

Today's Net, you know, a 1973 Buick, been doing all this work to refit it, reconfigure it and add things to it, but under the hood, it's still a 1973 Buick.

And we -- as we look at this situation, we see this incredible potential of not billions, but tens or hundreds of billions of connections to the Net, this need to revisit the architectural paradigms of it as we look forward. And we can't just effectively or incrementally enhance it at the base level; we need to look at something that we can put on top of it. And as Larry Peterson, right, the chair of research at Princeton, says, "The Internet has reached a plateau in terms of what it can do. The right thing to do is start over at another level."

And that's exactly what we think we need to explore as the next step in the evolution of the Internet.

Before we look forward, let's, again, just take a quick look backward. And the situation that we're envisioning, right, is one, where, as Vint described so eloquently, we had this set of networks, and they were all independent, not operational with each other, applications and services couldn't cross those networks very effectively. And instead, what Vint and Bob did was overlay the simple model of TCP, defined the notion of a packet. And it overlaid, it hid, it pushed down that complexity, and ignorance was bliss. Application developers could operate independent of knowing quite how the packet got from here to there, right, and the underlying network folks wouldn't necessarily be encumbered or struggling to deal with what the applications on top needed to require of the network itself.

And that was good enough for a very long time. But, perhaps not anymore.

And as we see the situation today, right, the Internet is growing and, essentially, replicating that network of networks and all of that underlying complexity and heterogeneity is reemerging. We have all types of networks seeing what we're trying to address the underlying network protocols, IPv6, Internet 2, trying to address the requirements of new usage models with things like multicast, trying to extend the range of the Internet with new media like, you know, 802.11 networks and, you know, the extension to mobile IP into the mobile domain. All of these things, right, are creating this wired and wired[?] disparity, heterogeneity and disparity of moving the Net forward today. And at the same time that the network is growing with this expansion and groaning over that, the applications are becoming more, not less, demanding as we bring applications like telephony or video streaming on it, they want more precise control, not more unpredictable characteristics coming back from the network. And the way a packet is routed, the way links are set up can have a make-or-break impact on the operation of some of those new services.

And seeing that picture, we think that once again the same paradigm that led to the creation of the initial Internet, it's time to rethink and replicate that exact architectural thinking for the situation of today. The solution is creating, right, an overlay network. Right, creating an overlay that allows us to again abstract and look past much of that complexity.

What Vint touched on earlier, he described the creation of it by superimposing this new layer, an overlay. We think that overlays are the best way to approach the situation of today. And with that in mind, our view of the future is is that we create what we'll call a computational services overlay, a new paradigm where we don't just look at end-to-end connectivity, but we create computing and storage resources that become an inherent part of the infrastructure of the network itself, right, where those infrastructure services that can then operate and understand the traffic patterns, the hardware utilization, link failure, and be able to adapt and respond to those different characteristics. Right, and presenting not network bandwidth, but services of the network resources that are available. And create a more comprehensive view of the instantaneous health of the Internet and be able to respond and address it.

Now, many things are going on in the Net today, so, Pat, how does this relate to things that are happening today? You know, Vint talked about IPv6, and we think v6 is wonderful, it improves the characteristics of the Net. But it could be years, if not decades, until it's put in place. And to be homogeneous again may never occur.

Right, we see things like Internet 2 and v4. You know, this complexity, we would overlay it with the computational services overlay, allowing us to, again, abstract and isolate many of the difficulties of that underlying network layer. And we have many other things going on, such as Web services, grid computing, distribution requirements. Those, then, would become services that could rely on that computational services overlay or applications that rely on that services overlay. And thus they could be abstracted again from the difficulties of the underlying network.

And once again, we replicate the paradigm of the TCP/IP protocol, but we do it at a higher, more abstract layer, and a much more capable layer.

Let's very quickly look back at the challenges that we were describing earlier when we talked about Vint: Capacity, reliability, security, accessibility, and regulatory. And while this notion of a services overlay, we think, is powerful, it's not a panacea. But does it address some of these challenges that we described earlier?

In terms of capacity, new services would run on top of this and it would be able to respond and address, right and move capacity and requirements and move resources to where it's required. Look at the underlying network traffic and be able to migrate its services to the areas that are demanding of the network resources that are available.

Right, if you look at an example like, maybe, France '98, 37 million page hits, right, looking at the soccer finals. Tremendous overload. The network would dynamically allocate resources to those domains of those flash mobs, right, and see, be able to allocate more of the network's resources to address those problems.

Reliability. The problems addressed in the network itself, right, where the network could understand its characteristics as failures and begin to self-diagnose, self-heal, and address and respond to those network characteristics.

Security. The network would have a global view of the network itself, and it would be able to see the transactions, the trends, the different attacks and services and viruses on the net and begin to self-respond to those and be able to deliver a more secure network inherently in the services that it offers.

It would be able to respond to the accessibility requirements. It would understand that this packet that's going to a remote village in India may take six more hours until the bus shows up to deliver it. It would be able to store and keep those things and stage the network delivery of that traffic appropriate to this increasing disparity of services.

It would also, as a network, be able to -- based on that visibility, be able to address to and respond to viruses and different geographic requirements.

The privacy and security laws in Europe might be different than those in the U.S. We can't embed those into underlying protocols. We need the network to be able to respond and address those different regulatory regimes.

The network must enable the market driving and creation of those regulatory frameworks.

Now, this sounds like pretty heady stuff, and particularly from a chip head company like Intel. What on earth is Intel talking about a global planetary service layer network? Don't you guys do chips? What gives us such confidence that we can talk about such bold views of what the network and society and geopolitical efforts would look like in the future on this network?

We're optimistic because we've been doing the core research, and we initiated a body of research that, much like the early Internet, it's gaining momentum. It's becoming viral and enthusiastic.

We've created a blueprint for what the Nets of the future might look like, and that effort is called PlanetLab*.

In Internet speak is rough consensus and running code. I don't care about your ideas you might have in the lab. Show them operating in the real Net. Show them really up and running. And that's exactly what we've done.

We've put up a prototype network and it's getting very exciting.

We started PlanetLab in 2002. We sort of forklifted the base services, put in place the first up with 100 machines across the globe at 41 sites in 7 countries. And after we got the basic infrastructure in place, services up and running, it generated a lot of enthusiasm. And today we're proud that we have over 440 nodes running at 194 sites and 22 countries. And this has been far more successful than we imagined. We had in the final 15 days in getting the slides ready for the keynote, we had many new entrants to PlanetLab. People are getting on board quickly.

And it's universities, research institutes saying wow, this is a great way for me to pursue my research interest. And now we have about 150 universities, the top universities around the globe jumping on board.

And the research interests of many corporations are becoming part of this as well, companies such as Hewlett-Packard*, one of the charter members in launching the PlanetLab with Intel. Others like AT&T*, NEC*, Google* and France Telecom*.

We're also seeing the research network efforts, the national research efforts like Internet 2, and Brazil with RNP* and China with CERNET*, they are jumping onboard as well and participating with PlanetLab.

What is PlanetLab? We're speaking to it. We're seeing this enthusiasm, but what really is it?

And today, you have a router in the network. Tomorrow you would have a computer delivering services in the network. And a PlanetLab node is essentially an Intel-based server running a virtual machine layer, some basic services that then allow the instantiation or participation of different services on those nodes, and those are then distributed globally across the Internet.

So a machine, piece of hardware, instead of a router it's now a computer. On top of that a virtual machine manager and base management services that can monitor the node in the network traffic and finally allocate resources to the services that would want to operate across it.

And then building on top of that, every service could then take a slice of one of these PlanetLab nodes and operate their service on top of it.

So today, when I run -- when I get my network service from an ISP, I'm essentially renting a slice of a whole bunch of Cisco* routers. In the future, when I get a PlanetLab service, I'd be getting a slice, a virtual machine of a whole bunch of PlanetLab servers that are built into the network.

And on top of those basic services, then we build more comprehensive services. Some of these would be more infrastructural in nature, like event processing, being able to notify when things occurred. Network mapping, being able to understand the network routes and topology. Distribute hash tables. Being able to look at and distribute objects across the entire New York.

Other things would be services that users or applications might build on top of like content distribution. And this whole notion of PlanetLab, like IP: simplistic, but yet very powerful.

And we have a class in Room 306 in our training sessions today that just addition into the real technical underpinnings of what PlanetLab really is and how one of these nodes really looks and operates.

All that sounds pretty cool, but does it work? Can you utilize it in some interesting way?

And one example I like to give is PHI or public health of the Internet. What this is a service a couple of grad students at Berkeley, Ryan Huebsch at UCB, Berkeley Lab, said once I've now put all of this plumbing in place with PlanetLab, I have nodes all across the globe, can I attack the problems of viruses? Can I look at virus propagation, and in literally, just a few man months, he built what we will argue is the most powerful tool to address a problem that has plagued us. Imagine with all the brilliant minds in this room we couldn't address the issue of viruses to date? Why is that? Why couldn't we solve the problem?

The problem is architectural. Putting the services layer in place, we think we now have a tool to fundamentally come back and look at the Net in a new way.

So let's take a look at PHI. And what we have here, we have a look, a perspective of, this is a live connection to the real PlanetLab network today. You know, these 400-plus network connections across the globe. And what we've been doing, and I'll start this up, is since IDF began, we have been monitoring attacks on the network. So we literally have many thousands of attacks that occur. And what PHI is now doing, it's taking a snort service, and this is one that looks at and has several thousand rules that looks at and characterizes attacks and is able to identify them. Badly formed addresses, repetitive addresses, other things like that, and is able to identify them.

And what you're seeing on the screen is literally the last 24 hours of network attacks, where they came from and where they're headed to.

And now what we can do -- and that's pretty interesting and we can zoom in on the geography. Let's go look at India. A little bit of activity coming in and out of India, but not too much. Let's check out Taiwan. Taiwan is pretty mellow.

But let's go to the U.S., and we see a tremendous amount of activity happening inside of the U.S. A lot of stuff happening on the east coast.

But now we cannot just look at this data but we can start to analyze the data. And let me click on forensics here. Because now we want to start looking at the network characteristics and try to understand the propagation of viruses, worms, and other things like that.

And this particular analysis we're doing here, this is data that was gathered over three months over the summer, and we were specifically looking at Code Red. And this is analyzing Code Red, and you see the cyclical natures where Code Red would propagate throughout the network, it would kill itself off at the end of the month, feigning that IT managers had it under control and guess what would happen? Poof, it would reappear the next month and the ongoing battle of dealing with this virus.

That's pretty cool. Let's go and look now, what we find is the top ten sources -- these are dynamically changing, but if you just look at the top ten sources for viruses, denial of services attacks, et cetera, what was seen by our analysis, if you just take care of the top ten, you've taken care of over 60 percent of all of the attacks.

So wow, you know, there's a real 90/10 rule here. Get those top ten and you take care of huge amounts of the overall traffic.

And now what I'll do next is I'll actually turn on a PHI filter. And you could imagine that what you'd now do is imagine that we took PHI from this little research project from a lab and said we're going to notify the firewalls of corporations of the IP addresses for the top ten current attackers on an ongoing basis. And they'd be dynamically updating their firewalls to just block those top ten. 60 percent of service attacks, just by adapting to the forensics we're showing here would be eliminated before they ever hit the corporate network. Just the top ten.

And it's things like this that we believe show the incredible power of the PlanetLab. Where we've been able to take that basic infrastructure of PlanetLab. Based on that basic infrastructure, literally a few man months of researcher work has created the most detailed analysis of worms and viruses of the network that's been ever done today, and this can easily be turned into about a real business, a real service that would dramatically improve corporations' ability to handle one of today's nightmares of the Net today: viruses and worms.

And again I emphasize, this is real. We are on the PlanetLab right now as we're doing this.

Let's take a second example. Researchers at Carnegie Mellon said, wow, this is pretty cool. We have this PlanetLab infrastructure. Hundreds of nodes scheduled around the world. Now it's pretty easy for me to go look at a problem like webcasting and be able to make part of that infrastructure useful for webcasting or streaming services.

So they used PlanetLab and created a media proxy service that allows and runs on top of PlanetLab.

And we think again, this is a blueprint for future commercial deployments. Think of this as PlanetLab based Akamai of the future.

And the idea is today if you were trying to webcast, and imagine Lord Browne was trying to speak to all of the BP employees, his network administrators would hate that day because they'd have to go reconfigure and statally set up this huge network to reach their tens of thousands of employees across the globe. And it's all static, can't change and when he's done with the keynote they rip all that out and go back to the normal network operation. Very intrusive and expensive.

Or you end up with a peer-to-peer operation where the network doesn't know any of that stuff and you end up with the picture that's shown on the graph here where you end up with multiple streams headed to common locations. How many streams do I have going to China, Taiwan or someplace else on the network? And this creates huge redundancy and inefficiency in the network itself because they're carrying all of this traffic but because of end-to-end transparency, the network can't respond or address or make itself more efficient against that workload.

What we're doing today is we're actually webcasting the keynote, so the whole time we've been talking today with Vint we have been webcasting the keynote across the PlanetLab network. And what you're seeing on the slide is a topology of the PlanetLab network and the source here is a webcasting node. It's then broadcasting over the network. You'll see there's a branch for the east coast universities. Obviously the nodes have found out that their nearest PlanetLab neighbor is nearby on the east coast so they've configured themselves as a network. Another branch up the West Coast. We also have branches that are international. So we have literally the network is dynamically configuring itself to optimize the delivery of the webcast of this keynote as we speak.

And now what we'd like to do today is take these 40 or so PlanetLab nodes that we have up and we're going to add a few more. And we have six monitors around the screen here and we're going to add six more nodes but for these nodes they're not aware of PlanetLab and don't have the media proxy service turned on.

So as we bring on these additional elements to the network, we're loading down that media server on the network today. And as we bring up one, two, three, four, you see the service degrade.

And in fact, this is exactly what we see on the Net today. More and more hits to a common website, more and more media streaming loads, it gets worse and worse and worse, and what eventually happens? It just becomes simply unusable.

And if you look at our topology here, as we're adding these nodes, you see them incrementally joining and putting more load on this media server that's on the Net today.

And as we watch here, frame rates are declining as we bring these up and it's getting jerkier and jerkier as we add more and more load to the network.

Now what we'd like to do is turn on PlanetLab. So now what we're going to do is add another PlanetLab node, all right. And this -- the node is right here on stage. So the man behind the curtain, right, is literally turning on another PlanetLab node. You'll see the PlanetLab node enter the network. And this PlanetLab node happens to be on the local subnet. And as he enters the network, what you will see is, he'll join, he'll attach to the source. And all of these other clients that we've put on that Web server eventually will say, oh, I'm getting high frame rate drops, I'm getting lots of packet losses. The network is long to get to that location. And they're going to reallocate themselves, they're going to move themselves to the PlanetLab node.

So you see the orange dot has come on here. Right, you've seen all of the other network nodes, the other six that we have around the screen here are reattaching themselves. And voila, the video quality has improved. The network has, you know, adapted to the network load of a media service, built an effective network across the subnet, eliminated huge amounts of traffic going over the network and dramatically reduced the load on the Web server we had, you know, starting with.

You can think about this as maybe adaptive Akamai, done by a few researchers at Carnegie-Mellon who just built on and utilized that service. And, you know, as you look at the picture on the slide now, what you see is, we've essentially created with this adaptative media service, eliminated six parallel streams, dropped it to one, to be able to create this efficient network as well, adaptive Akamai. I think there's a business opportunity here in the future.

PlanetLab -- that's what the topic is -- PlanetLab has started to get a lot of attention, a lot of interest. We've seen everything from Wall Street Journal to Business Week starting to look at this and say, "Something interesting is going on in this space called PlanetLab."

And today, in fact, based on the research that we've done, the interest that we're getting, just like the Internet, where we had this great enthusiasm and an interest in the research domain, it spilled over into the commercial domain. And today I'm very happy to announce that Hewlett-Packard is joining with Intel in the pursuit of commercialization of PlanetLab services.

The network, the architecture, right, this initial infrastructure we've put in place is so promising and so interesting, we believe it's time to begin to look at commercializing services that are built on that network. And I'm very happy, Shane Robison, a friend and comrade, the CTO of Hewlett-Packard's quote is listed here, as they are joining with us to accomplish that commercialization of PlanetLab.

We're also very happy today to announce that PBS, the Public Broadcasting Service, very technology savvy, very interested in how they can use technology to further the reach of their content and services, are working with Hewlett-Packard and with Intel to begin using the service that we just showed in a commercial environment and begin to deliver to deliver high-definition TV of their service using that architecture. We believe these are just the first of many such opportunities that will emerge.

Our vision for the new Net, based on this overlay, this computational services overlay, this architecture that we've been researching with PlanetLab, this tremendous enthusiasm and interest, is that we transform the Net. It moves from packet communication to a service hosting platform. We go from having routers to having full-service capable, that's communications, compute, and storage, inside of the network. And instead of just end-to-end transparency, we've had to put more edge services into it, then in the future we end up with these nodes literally everywhere in the network.

We would encourage you to join PlanetLab. We think something big is starting here. We think you have an opportunity to participate with us in the Internet transformation. We think the work we're doing today is laying the foundation for the Internet of tomorrow.

Thank you very much.

About Intel
Intel (NASDAQ: INTC), the world leader in silicon innovation, develops technologies, products and initiatives to continually advance how people work and live. Additional information about Intel is available at www.intel.com/pressroom and blogs.intel.com.

* Other names and brands may be claimed as the property of others.