Intel Developer Forum, Spring 2001

Pat Gelsinger
San Jose, Calif., USA
February 28, 2001

ANNOUNCER: Ladies and gentlemen, please welcome vice president and chief technology officer, Intel Architecture Group, Pat Gelsinger.

(Applause.)

PAT GELSINGER: Good morning. Great to be with you again today. Thank you very much. Contrary to popular belief, I am a farm boy. I do know the difference between a mare and a stud; so next time I go see Pentium, I'm going to make sure it's a prince or princess and I'll give you a full report at the fall IDF.

(Laughter.)

PAT GELSINGER: How many of you have heard about peer-to-peer computing? You guys out there? Have you guys been on a desert island for the last decade or year at least? peer-to-peer computing is very hot and I want to talk about what we see as the next steps required to take peer-to-peer from an interesting fad to the next generation computing architecture for the entire industry.

Our summary agenda for today is peer-to-peer is very hot. We're going to talk about some of the things we've seen making it very, very hot.

We're going to talk about the progress that the industry is making, and finally, what we see as the next steps to make peer-to-peer computing this broad computing infrastructure for the next generation of the Internet's evolution.

In the fall IDF I showed you this slide, and I made a bold suggestion. I said peer-to-peer computing could be the next revolution in computing for the entire industry. And I pointed to Napster*, Gnutella*, FreeNet* as the triggers, those that changed our perspective and gave us the first view of peer-to-peer computing and the potential it might have. And this pronouncement at the time seemed a bit bold, and I'll suggest to you today that our very optimistic and bold estimations at the time were yet conservative compared to what we've seen occur.

We've gone from almost nowhere to everywhere in six months' time. Looking at some of the data and announcements, you see it literally everywhere in the industry. We see announcement after announcement of different companies, of potential impacts of peer-to-peer computing, new products being announced, new services emerging. And between every magazine, newspaper, and legal journal on the planet, you see something about peer-to-peer computing. And even just last week, we saw Sun announce JXTA, their proprietary island and attempt to control the peer-to-peer community. Everybody is getting in the act. peer-to-peer is very, very hot.

Some of the early data that we're seeing, we see two particular applications that have been very stunning in the industry. We've seen applications of file sharing, music sharing. We see Napster has over 62 million users today. That is not including the many Napster look?alikes and other equivalent technologies that are emerging very, very rapidly.

The second killer application today is instant messaging and over 200 million instant message users today. These are huge numbers. Various predictions are emerging showing peer-to-peer computing and broadband technologies in particular will enable a broad new set of applications, such as virtual private Webs, photo, calendar sharing, et cetera.

We've also seen that the number of companies in the peer-to-peer computing space has increased very rapidly. When I talked to you in the fall about peer-to-peer computing, we had somewhere around 60 companies we could identify in this space. Today we see over 150 companies covering a very wide variety of aspects, as shown here.

We see our first books emerging, peer-to-peer Computing, How to on Capitalize Disruptive Technologies. We also have first drafts that we can show you in the demo showcase of Intel's first book on peer-to-peer computing as well. Very quickly, everybody is trying to understand this trend and find out where it's going. peer-to-peer computing is very hot.

I was very happy in the fall IDF to announce the formation of the peer-to-peer Working Group. And at the time, it was a loose announcement of our intention to create such a group. Today I'm very happy that we have, in fact, created the peer-to-peer Working Group. It's organization, structure, and processes are defined. We have the officers elected for the peer-to-peer Working Group, and those officers are Brian Morrow from Endeavors Technology, the steering committee chairperson; Andrew Grimshaw from applied meta computing, Jeffrey Kay from Engenia, Andrew Chien from Entropia and, Tim Matson from Intel form the steering committee for the peer-to-peer Working Group and they were elected at the most recent meeting.

Finally, the technical architectural council chairperson is Bob Knighten from Intel. And that leads to the next bullet, which is the goal of the peer-to-peer Working Group to establish an interoperable framework for peer-to-peer applications that's being driven by proposals being received from several in the industry for recommended practices or the framework by which we want to stitch together a broad peer-to-peer platform. And our next meeting for the peer-to-peer Working Group is in May.

These are all steps in the right direction. And we're beginning to see the proof points emerge for peer-to-peer applications to be broad, and more significant than even we envisioned.

In the fall IDF we showed this picture and we had categorized these early applications into these five usage models: Collaboration, edge services, resource sharing, universal file sharing, and intelligent agents. Let's just look at a few proof points against this model and see how we're doing and some of the progress that we as an industry are making in this area.

Since our last meeting, we had Groove Networks announce the Groove* platform late in the year, and is rapidly developing its platform of services that they are bringing different application developers and business partners to with over 200 of those developing, tools, application services and end user applications. Finding customers in government and pharmaceuticals; finding an international presence with over a third of their downloads coming from non?U.S. locations, clearly a worldwide trend.

Next example of collaboration would be Endeavor's technology. What I'd like to do is give you a brief demonstration of Endeavors and I'd like to introduce Brian Morrow, the president and COO of Endeavors. And you're also the --

BRIAN: Chair of the steering committee of the peer-to-peer Working Group. I'm a busy guy these days.

PAT GELSINGER: Can you tell us a little bit about Endeavors Technology?

BRIAN: Sure. This morning we released the freeware download version of the new version of our Magi* technology, much of which we'll be showing today. The MAGI technology puts a thin server on every network-connected device. The client and server sides work together like two sides of the same coin to create a peer, and the peers on every device work with every other device to interoperate, allow people to find and share information from all edges of the network inside an organization, no matter what they are.

PAT GELSINGER: Those devices, that's clients, it's servers.

BRIAN: Actually we turn everything into a server, but it can be laptops, desktops, and today we're going to show it on palm tops.

PAT GELSINGER: Very good. And to help us with our demonstration we have Sharad Garg from the Intel peer-to-peer lab. Hi, Sharad. Thanks for joining us. What are we going to demo today?

SHARAD: This shows available information on any device, anyplace, anytime. You've got to sign a legal document in a peer-to-peer collaborative fashion, share the document with Brian. He's going to sign it using his iPAQ* and share with you, and you will sign it and share back with me. The iPAQ uses Intel® StrongARM™ processor and is using 822.11 wireless protocol.

BRIAN: I hope this legal document isn't too binding.

PAT GELSINGER: Let's find out what we're going to sign here. Is this going to be a big check you were going to send to my account?

BRIAN: I was hoping it would go the other way. I'm watching my collaborative space here on the iPAQ and here's the document.

So I'll just pull that up here. Perhaps you could hold that for me. I just don't have three hands today. And I see it's got a place for me to sign. So I'll just sign my name. Sign away my life here. And then I will pull up the collaborative space. Okay.

PAT GELSINGER: In front of all your friends here, I see your manual dexterity with iPAQs, your demo skills. This is great.

BRIAN: It's very kind of you. I practiced all last evening.

I'm now sharing the file, and let's see how I can share it with. You're here on my list, so I'll just share that with you.

And in a moment I'll get a message that hopefully it's been shared successfully.

PAT GELSINGER: Okay.

BRIAN: I see the world turning.

PAT GELSINGER: Yep, and on my iPAQ now it's shown up here, so I see Brian, iPAQ, and I'll just click on this and, voila, I see your signature here so I'm going to add my John Hancock to this.

See that. Mine is pretty ugly, isn't it.

(Laughter.)

PAT GELSINGER: And then I'm going to share this. Click "okay" here. Explore, go back to inbox. Okay. Got it, Sharad?

SHARAD: Yeah, here.

BRIAN: This is exciting here.

SHARAD: I received the file with both of your signatures.

PAT GELSINGER: Okay?

BRIAN: Sharad?

SHARAD: Yeah.

PAT GELSINGER: Okay. Very good. So we see the completed document signed by both of us.

Brian, this is a nice scenario, we're seeing the shared collaborative space, wired, wireless devices. But what are the real-world applications of such a capability?

BRIAN: We're working with one of our value?added resellers right now to use these devices in a hospital setting. They're putting Compaq iPAQ in every doctor's pocket. As the doctor travels around the hospital, he'll be able to use the iPAQ to call up the patient records, to write prescriptions, and to contact other doctors directly who are also carrying iPAQs.

So the peer-to-peer network will be iPAQ to iPAQ all around the hospital.

PAT GELSINGER: So as well as the PCs, the servers.

BRIAN: Everything becomes a server, including the iPAQs, so we're dealing with the patient record servers as well as each doctor being a server unto himself.

PAT GELSINGER: Okay. Now we'll have new excuses for doctors having bad signatures.

BRIAN: Great. Thanks, Pat.

PAT GELSINGER: Thank you very much, Brian. Thanks, Sharad. (Applause.)

PAT GELSINGER: When we talked in August, I described our plans for some internal uses and trials at Intel. And we launched what we called our edge services trial, our share?and?learn network inside of Intel.

We had the objective of delivering media training, an application we said was too expensive to do in a client/server way but one we thought might be appropriate in peer-to-peer. Since then we launched that trial inside of Intel, and the results have been very good. We took international training courses that we put online, 60 different modules of 10 to 20 megabytes each. We made those available online. We have over 2000 participants in this internal trial of doing many file transfers, many training sessions across 50 different sites at Intel.

The results have been excellent. We've had over 5 to 6X improvement in the file access times by users, over 80 percent of the accesses are delivered in a peer-to-peer manner as opposed to the original source. And as we developed and refined the technology for this trial, we went from 47 percent to almost 90 percent of the traffic being moved off the WAN and onto the local?area network.

This trial that we've run has been very successful inside of the company. We're now going to turn this over into a production application Q2 of this year, making it a available to all 85,000 Intel users. In addition to that, we're going to begin working with other IT companies and ask them to take this code, this learning, and begin to deploy it inside of those major enterprises as well.

Another example is compute sharing or collaborative resource sharing, compute sharing. And in this case, we like to demonstrate Livermore Software Technology. Now, this technology is the world leader in crash?test simulation. Literally every car company on the planet takes advantage of the software done by this company. It's a very compute?intensive application, and what we've done is, working with them, we've added a peer-to-peer layer to this application. And we'd like to give you a demonstration of that today.

Let me point out this demonstration is in production use at Ford Motor Company today. This is how they are doing crash test simulation.

To help me with this we have Bob Kuhn from Intel's KAI division. Hi, Bob. Great to have you.

We also have from Livermore Software Technology Nielen Stander. And also Tom Owens from our demo group.

So, Bob, tell us about the demonstration you have.

BOB: What we have here are dual Pentium® III workstations running at 1 GHz. The far one is running Windows* 2000 pro and this one is running Linux*.

First let's start by taking a look at LS-DYNA here. You see an LS-DYNA simulation of a Ford Explorer frontal impact. This is very compute intensive. This runs on processors for several hours.

LS-OPT, on the other hand, is an application for optimizing car crashworthiness and what it does is distribute processing to a whole new level.

With the Visteon instrument panel that you see here, we launched no less than 70 LS-DYNA runs. To make that kind of processing power available we use a KAI application distributor middleware for peer-to-peer computing, that allows it to be spread across a cluster or network of workstations.

Now let me talk about peer-to-peer computing as a collaborative design process. What we're going to do is run through a scenario between Tom and Nielen as automotive engineers. What happens is Nielen starts the process by describing a problem. He sends an e?mail to Tom, and Tom replies that looks good and Nielen launches a run.

Then Nielen brings up his application monitor. Tom also can watch the run in progress, make comments on it. Tom finds a problem with the run. That problem is an oscillation, and he sends a message back to Nielen. Then he says, "I think I can do better. I can pose another run, run 1.1." Tom launches that run and now we have two runs simultaneously sharing resources.

In the meantime, Nielen is looking at Tom's run. He sees Tom's run really is better. What I'm going to do is shut down my run. He does that, and the resources are now all applied to Tom's run.

PAT GELSINGER: So what we've seen here, Bob, is a very compute intensive application requiring many hours, and now we need to run that for a real product, 10, 70 times in this case that you said, so we now are distributing that across many resources. But we have to collaborate. So we've added both the compute sharing and the collaborative element on top of that allowing the engineers to work together to come up with the right design point.

BOB: That's right, Pat.

PAT GELSINGER: Thank you very much, Bob.

BOB: My pleasure.

(Applause.)

PAT GELSINGER: We believe this opportunity of resource sharing, of taking advantage of peer-to-peer computing networks, is pretty stunning. In fact, we are off developing the world's largest virtual supercomputer that's ever been assembled. And in a couple of months we'll be able to describe very specifically to you the applications and the plans to make this available to the industry.

And if we compare the potential of peer-to-peer computing against today's largest supercomputers, the numbers are pretty stunning. A five teraflop sustained supercomputer is the largest built on the planet today; it costs about $100 million to assemble such a configuration. For a fraction of that cost in terms of working with, facilitating the industry, creating the software components, and garnering the industry's resources and the end?user compute nodes, we see that we can do the equivalent for less than a million dollars and be about an order of magnitude greater in peak performance possible by such a virtual peer-to-peer supercomputer. And we're very excited for the possibility that such a supercomputing capability will enable in a peer-to-peer fashion.

We've also been working with the industry in understanding what the potential marketing requirements are for peer-to-peer. We ran about 13 focus groups to both consumers and IT users, small and large businesses, and we found some very interesting results from these focus groups that I'd like to share with you today.

The first is that where peer-to-peer is in applications today, inside of corporations and small businesses, it's a very positive experience. They're seeing financial services, academia, other industries are beginning to jump on board. It's seen very positively, very enthusiastically. The value proposition is seen as compelling, both to the IT manager and the resources they'll be able to harness as well as the new applications available to the end-user. And they see the need for peer-to-peer moving from narrow applications to a broad set of segments of their usages.

But their general expectations in the future are peer-to-peer computing becomes very large and significant to IT; however, there are some requirements to enable that to occur. It must be compatible with the current operating environments that they have, and there are critical tools and technologies, specifically in the area of manageability and security, that are essential to enabling a broad deployment of peer-to-peer.

This leads us exactly to what we think is the key steps that the industry must take to deliver the tremendous promise, the overwhelming benefits of peer-to-peer computing. We have to focus on interoperability and compatibility between the applications and the infrastructure that's emerging.

Let's look at the situation we're in today. We call this islands of incompatible technology. These applications in and of themselves are very good and very exciting, but in the collaboration area, one app is not able to collaborate with another collaborative app. They're incompatible islands.

Furthermore, as we look at those applications versus other applications, such as edge services or resource sharing, one is not able to interoperate with the other. In the example we just completed with Livermore Software Technology, you saw the powerful benefits of combining resource sharing with collaboration to allow not only massive quantities of compute resources to be deployed but the users themselves, in this case the engineers, to be able to collaborate across those environments.

Furthermore, based on our IT discussions, it's very clear that these applications must be delivered and enabled on the infrastructure that IT and corporations and users have put in place today. It has to work with, take advantage of the manageability, security, and underlying network and OS infrastructure. We need to move past islands of incompatibility to a broad platform of compatibility and interoperability.

We have a vision, if we were across the street in the Martin Luther King building, we might say we have a dream of what we're trying to create, and that's a set of services. A compatible, cross-platform infrastructure that's going to enable a significant step up in the potential of peer-to-peer computing. These services would be available in a common target for application and industry development. That based on these services, IT could have confidence in the manageability, the security, and other capabilities of that infrastructure.

And finally, it will facilitate interoperability between those applications. Looking at the stack, we see a set of services that we've been looking at, refining, discussing with the industry. An example might be availability. When I have a network of peer-to-peer devices, sometimes they're not always connected. How am I going to deliver a reliable service when I have occasionally connected peers? We need a technology to address that.

How am I going to be able to identify and find communities of users and deliver those in a secure environment? Those are some of the problems that need to be solved and done in an industry-compatible, interoperable way.

This is a very big task. You're going to hear Bob Knighten in the peer-to-peer track today discuss more specifically what goes into each of those layers but at this point, we say there's a couple of examples that we think are those highest priority and ones we first want to focus on. Specifically in the communications areas, the idea of common protocols that are language and platform independent, probably on SOAP as best example. Also see problems in the area of addressability. We see opaque areas of the Internet where peer-to-peer application can't work behind the NAT or firewall. Critical issues for us to solve.

In the peer-to-peer track you're going to see Microsoft speak very specifically about their technologies and plans in these areas.

Availability, how we replicate for occasionally connected devices, how we're going to fail over and have guaranteed message queuing. At the security level, we see security is required at each of these levels, but in and of themselves we need a consistent policy for how we authenticate, authorize, and have integrity and privacy delivered on the peer-to-peer infrastructure.

And finally, how we're going to be able to identify different users, how they're going to be able to be there persistently across different network connections, across different devices and different sessions.

This is a very big task, one that we see as taking several years to deliver, but one that we see through these road map elements that we're going to start and deliver very specific technologies in the next year to start solving these problems, working with the industry, through the peer-to-peer Working Group to solve.

I'm happy to announce the Intel Marquam program, which is not an architecture, not a product, but it's our program that we're dedicating resources with the industry to deliver. Specifically it's developing the technologies we just described to deliver on that industry vision. It's then driving the marketing programs to create and make peer-to-peer computing broadly accepted in the industry to both consumers and to businesses and IT accounts.

And finally, through things like our Intel Capital investments, the many marketing programs, catalyzing the industry. We believe peer-to-peer computing is the next revolution in the Internet, and we will, through the Marquam program, drive, lead, create, and make that transition to peer-to-peer computing happen.

What you need to do simply is get started. peer-to-peer computing is very hot. We invite and encourage you to invest heavily, look seriously at the potential for it. Through the peer-to-peer Working Group we ask for your participation. And finally, get religion. Through integration and interoperability, that's how we're going to move peer-to-peer from narrow applications to a broad platform of services and next generation capabilities. We want to work with you. We want to make it happen. peer-to-peer computing is the next revolution. Let's join together and make it occur.

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.