Oracle OpenWorld 2001

Craig Barrett
San Francisco, Calif., USA
December 3, 2001

(Video playing.)

DAVID: This is Spinal Tap. Thank you all very much. A very special thank you to OpenWorld for opening the world a bit more so I can squeeze in.

What am I doing here? Is that what you're asking yourselves? Some of you are muttering, others of you opening your mouths over and over again. Computers and me, after all, what's the connection? I'm a hard-working, meat-and-potatoes rock and roller. How does my life intersect with the world of bits, bytes and Web sites? Rather ask how does it not.

I remember being eight years old, sitting at the breakfast table when my dad showed me this story in the newspaper. MIT had just unveiled Whirlwind, its brand?new, highly sophisticated computer. It contained, as you probably all know, 4200 vacuum tubes, 14,800 diodes, miles of paper and magnetic tape and required floor space in excess of 3100 feet.

Now, my dad was a bit of a visionary. He ran a modest traveling business repairing luggage at various resort locations. An old fashioned craftsman he was. He could fix the handle on a Gladstone bag faster than you could say Joan Robinson. But he was fascinated with this brave new world he could see birthing itself all over the kitchen table. "David, he said, you watch. These computer things will become more sophisticated, more versatile, smarter. And some day, scientists will find a way to make them smaller. So small, they might fit in a briefcase or other carrying device. And when the handles on those cases break, and they will, I'll be there, Johnny on the spot, to fix them."

(Applause.)

(Laughter.)

DAVID: Yes, like I said, he was visionary, but he knew his limitations.

(Laughter.)

DAVID: But I thought he was mad, you know. I was only eight years old, but I knew a bit about Babbage's difference engine, and I saw the stats on ENIAC, the 1946 machine built by John Mauchly and J. Presper Eckert. And that baby took up a mere 1,000 feet of floor space five years before, a third of the size of the Whirlwind. Obviously, the trend was going to be bigger rather than smaller.

Fast forward 20 years. By the early 1970s, the computer had learned to play chess and run assembly lines but it wanted more. It wanted to make music, create poetry, produce miraculous films and video effects and above all to dance. Well, most of these have come true, hasn't it, apart from the dancing bit.

In the mid '80s, I utilized much of this technology in the service of my muse. I had the dream of a cyborgian collaboration, man and machine co-creating a new hybrid of music, as free and human as an Irish reel. It was theoretically, the best of both worlds. Unfortunately, I was using a Commodore 64. I'm still waiting for it to print out.

Now, of course, things are moving much swifter thanks, in part, to the miraculous Intel processor and its successive successors. And my dad's vision has come true, too. Computers have become much more portable, and the handles still break occasionally, as do those retractable cupholders that come built into some of the better consoles.

And great visions like those that disturb your dreams and make-up sessions here at Oracle OpenWorld are really just the cumulative vision of all visionaries, from Charles Babbage to my father, a humble luggage repairman. Thank you. Now, let's rock and roll.

(Applause.)

DAVID: One, two, one, two, three. (starts playing the guitar) That can't be right. Excuse me. Can I get a techie here? What the bloody hell? Can I get a guitar tech? I'm supposed to be at 11. It's only going up to about three, I think. Can anyone help me here?

CRAIG BARRETT: David.

DAVID: Bloody hell.

CRAIG BARRETT: It's not working. They can't hear you out here.

DAVID: And you are?

CRAIG BARRETT: I'm your boss.

DAVID: Okay. I was going to say you're a bit overdressed for a tech.

CRAIG BARRETT: No tie. But hey, you're just not making it.

DAVID: What are we going to do?

CRAIG BARRETT: You need a little help in your deployment, I think. I brought a consultant out to help you. This is Jude over here.

DAVID: Hey, Jude. Bet you never heard that before. What's the problem, do you think?

CRAIG BARRETT: I think the problem is you need more processing.

DAVID: Processing, huh?

CRAIG BARRETT: Processing.

JUDE: Dude, you've got to go macro.

(Playing guitar.)

DAVID: Well, then to recap ??

(Applause.)

DAVID: You're saying that's macro.

CRAIG BARRETT: You got it.

DAVID: I think I need a bit more consulting. I'll be back. Thank you, Craig.

CRAIG BARRETT: All right. I'll take over from here.

(Applause.)

CRAIG BARRETT: My job out here today is to tell you how to take your enterprise data centers to an eleven. If you don't know what that means, you have to go out and buy the Spinal Tap DVD tomorrow and look at it, or tonight.

We're going to talk about macroprocessing and talk a little bit about the future of computing. The agenda I have is very simple. Why macroprocessing? We'll talk a little bit about the economics it brings you, the open industry standards that it brings you, the ability of the whole industry to innovate around open platforms.

We'll show you a few examples, basically the examples under the title of a backstage pass. We'll go back through a few examples and look at what's going on in the industry, and then we'll come back and look at a little bit of the future.

Before we talk about that, I want to just spend a few minutes and talk about the economics of computing and the Internet as it's going forward today. And one of the main issues is we've all had the excitement of the dot-com craze and crash. We've had the computer industry proliferating for the last five years but the industry is in a downturn now for everyone.

But if you look at the history of economics, you'll see this is a typical phenomenon. Every time a new innovative technology comes along, whether railroads, the automobile industry, television--you pick it--there's always an enabling technology that comes off first, a rational exuberance, to quote Allen Greenspan, that follows, and then usually a crash where people become convinced once again that it's real products, real technology, real revenues, real profits that make companies successful, and then you see the golden age of the industry following that.

I think we've passed the rational exuberance for our industry and are now in the golden age. The golden are of computing and communicating are still in front of us. So the growth we're going to experience over the next 10, 20, 30 years is the exciting part, and that's the basis for the conversation today. What's going to be the baseline, what sort of computational capability will you need for that?

Now, why macroprocessing? What does that term mean? It really means bringing the economics and the capability of the microprocessor to the big data center application. It's really talking about innovation. Not one company's innovation, but the ability of a whole industry to innovate around a common architecture, modular standards, the capability to build up a system as large as you need, but to bring the volume economics, the price, and the performance leadership, both price leadership and performance leadership, and then the ratio gets even better, to bring that capability we've seen for the microprocessor and what it's done for the desktop, bring that capability to the data center. That's really what the phrase macroprocessing is all about.

Now, if we look at what we could talk about from a deployment standpoint, if you look at the slide behind me, you'll see there are two categories, and I really tried to differentiate between the proprietary approach and the macroprocessing approach. I'm not going to spend any time on the left-hand side. Scott McNealy will take care of that on Thursday. I'm here to talk about the right-hand side of the slide today.

(Laughter.)

(Applause.)

CRAIG BARRETT: We're here to talk about distributed processing, modular standards, hardware/software available from a wide variety of providers, competitive innovation that takes place, incremental growth, adding just as much capacity as you need with standard modular building blocks, very cost effective solutions. This is what gives you maximum scalability, efficient scalability. And from this kind of a scientific standpoint, you might refer to this as macroprocessing being the elegant solution.

The elegant solution from the standpoint of adding just enough capacity to meet your needs; not having to change hardware, change databases, but just adding incremental capacity. So macroprocessing is really the elegant design for the future. And it's dealing again with volume economics, easy to deploy, easy to manage, reliable, certified installations and certified configurations, and all of the good things you would ever want.

Now, does this work in practice? And I think that's the question we are all asking. Can you really build these modular systems up into big reliable data center applications? And what we want to do is take a few minutes to look at a couple of examples, and I have four simple demos -- actually, two videos and two demos, that we want to go through to show you where macroprocessing is working. We'll try to run the gamut of companies from extremes of Oracle to scientific instrumentation to computer manufacturing, and to the Federal Aviation Administration. Everything from government to manufacturing to scientific exploration.

The first one I want to really look at is Oracle. We've been working with Oracle for a number of years. You just listened to Chuck give a presentation, and he did mention the close cooperation between Intel and Oracle.

One of the areas we've been cooperating very much with Oracle over the last few years is the convergence of voice and data. We all know that computing has been digital forever, but the recognition is coming more and more real each day that voice is digital as well. Anything you can digitalize, it's easier, cheaper to do that. So we're going off analog and going to digital system. With the convergence of voice and data, you only need one network, one backbone and one standard set of compute elements to handle the two.

What we've been doing is working with Oracle, some of their application servers, 9i RAC, 9i AS, for universal messaging systems and trying to bring that convergence of voice and data to the average consumer in a very reliable, unbreakable, economic fashion.

CHUCK ROZWAT: Great to be back up.

CRAIG BARRETT: I hope that little noise in the opening didn't bother you.

CHUCK ROZWAT: I was impressed. It woke me right up.

CRAIG BARRETT: We've got a great cooperation with you. Why don't you tell the audience a little about the work Oracle is doing.

CHUCK ROZWAT: You're right, the cooperation between Oracle and Intel has been great, particularly on this project. So we've put together a unified messaging project where we've taken the skills of Intel, the skills of Oracle, put our best skills together, and come up with a system that we're very close to going to market with. And it's not just a research project. We're actually using it inside Oracle. So we're using unified messaging for thousands of Oracle employees, and we're looking forward to rolling it out across the whole company.

CRAIG BARRETT: This is basically a combination of voice and data, putting it all on a common system, common database and you can get at it any old which way?

CHUCK ROZWAT: Absolutely, from any device you happen to be using. And we've learned a lot from Intel about communications, about voice, and we've been able to add the scalable software that Oracle does best to the solution.

As a matter of fact, we're looking forward to going beyond just unified messaging, being able to apply this to the whole e-Business suite so all of our applications can have this capability. So it's worked great.

CRAIG BARRETT: So that basically means we're not going to be able to go out and play golf and get away from business anymore?

CHUCK ROZWAT: No escaping anymore.

CRAIG BARRETT: I've got it. We're going to do a demo on this system from Oracle. Kit has taken the opportunity to tap into the Oracle database. Why don't you show us what you have.

KIT: It's a thrill to be talking about this, running on the building-blocks, front to back and this is live. We're going to surf in from the Internet. This is a URL we're going to pull up.

CHUCK ROZWAT: So you can get to my Web mail account over the Internet.

KIT: Yeah, it's like having the keys to dad's car right now because I have the password. Let me log in.

CRAIG BARRETT: You did give him your password?

CHUCK ROZWAT: Yeah. It's a little dangerous.

KIT: It might be wise to change it. We have Chuck's inbox. We have several message types. Benefit to the user is you now aggregate all of your message types in one location. You can see phone numbers for voice mails, you've got e-mails in here, it offers greater flexibility and management of your messages, you can access through any point, a land line telephone or wireless telephone, a PC or wireless PDA.

The IT folks love this, because it has tremendous savings when we use Intel's compute model, and we mix and control to bring everything together.

CHUCK ROZWAT: I've been using the system and it's great to have voice mail, e-mail, faxes all come to one place. You don't have to be searching, using different models, different equipment to actually get to my messages. One place, there it is.

KIT: Want to have a listen?

CRAIG BARRETT: Yeah. Pick out a voice mail and let's try that first.

KIT: Let's see who has been trying to get Chuck while he's been preparing for Oracle OpenWorld. I'm clicking on his voice mail and there's a download icon. I'm going to download this voice mail and play it with a wave file.

(Playing voice mail from Larry Ellison: "Intel and your team has been doing a great job. Keep up the good work.")

CHUCK ROZWAT: If you couldn't hear that, that was a message from Larry.

CRAIG BARRETT: That was your performance review, wasn't it?

CHUCK ROZWAT: The good news is he used the word "great" in there, so I'll take it.

CRAIG BARRETT: Can we do the flip side? Let's access this not with a computer but with a handset and see if we can get e-mail done over voice.

KIT: Okay. We're all used to seeing e-mail, so what I did is pop open one of your e-mails here and then I have a telephone, wireless. I'm going to go ahead and actually dial up Oracle's voice server that's going to read this e-mail. So let me just dial up here.

CHUCK ROZWAT: And again, this is running right off the live system that we're using at Oracle in our data center.

>>: Oracle e-mail. You have no new messages and six old messages. First old message from Denise sent today.

(Playing message.)

CRAIG BARRETT: Denise has got kind of a masculine voice, don't you think?

(Laughter.)

CRAIG BARRETT: Maybe your e-mail system could do name recognition and have a female voice and a male voice.

CHUCK ROZWAT: I think we need to do a little work on that.

CRAIG BARRETT: I think this is a great application of the ability to stay in contact, either monitor contact or voice contact, voice/data convergence, database, way to go in the future.

KIT: And this isn't it. The future brings a lot more function elements and applications built upon Intel's building blocks.

CRAIG BARRETT: Great.

CHUCK ROZWAT: Thanks very much.

(Applause.)

CRAIG BARRETT: I want to give you two other quick examples, one about Dell. Everybody knows the Dell computer model in terms of build to order, no inventory. It's interesting to see that Dell is, in fact, moving their basic IT infrastructure to an Intel architecture base running Oracle9i RAC. Let's watch a quick video to see what they have to say about this combination of technologies.

(video playing)

VJ: Welcome to Texas! We're here at Roundrock 5. Home of Dell's I.T. labs. Inside they've engineered a phenomenal volume-based solution for running Oracle 9-I rack. I got my backstage pass… let's go inside!

JAMES: Cash fusion and real application clustering is a real quantum leap in data base technology.

SUMANT: Building highly scalable databases for a growing company such as Dell, is a challenge.

DAVE: Before 9-I, to build a large database required buying very expensive proprietary hardware.

JAMES: The traditional model of deploying Oracle technology has been buying as many processors as possible on a given server to scale up.

DAVE: Where as in the past if you wanted to increase the capacity of your database, you had to add a whole separate server and migrate your database over.

BRENT: In traditional large SMP systems, sure, you can add CPU's to scale the system up to point, but then you have got all of your eggs in one basket, if you will, and if the system goes down, you're sitting dead in the water - business is essentially stopped.

JAMES: But if you have four, four-processor servers, and they're clustered together, it doesn't matter if one of those fails, the other servers in that cluster will automatically take over.

SUMANT: To support mission critical global applications in a company such as Dell, you need to have very highly available databases.

JAMES: High availability can be achieved with higher performance by using smaller four-processor servers clustered together.

SUMANT: Clusters of Intel-based, small servers at Dell will help us build large complex and more scalable database solutions for our own business.

DAVE: We're achieving the same or better performance for typically a fraction of the price.

JAMES: We can actually lower cost from both a hardware perspective and also a management perspective.

(video ends)

CRAIG BARRETT: That's one, I think very classic example. Dell is classic of a manufacturing machine for building real time, build to order. If the system goes down, they go down so they need an absolutely reliable mission critical application in place.

Another example of a mission critical application in place is the FAA, Federal Aviation Administration. You might not know it but there are about 50,000 flights throughout the day. At any point in time, there's about 4,000 to 6,000 airplanes in the air that the FAA is taking note of. After the incident of 9/11, it's absolutely essential to know where every plane is at every time.

Let's hear what the FAA has to say about clustering technology and how they manage our air space real time.

(Video playing.)

Welcome to Washington D.C. Today we've got a backstage pass for the Federal Aviation Administration's Air Traffic Control Command Center.

Now at any point in time there are literally thousands of airplanes in our nation's skies. Managing this air space, keeping flights moving efficiently and informing airports and airlines of minute by minute changes is a task that requires efficiency and reliability from both the command center personnel and its servers.

Hi, I'm Tim Grovac, I'm the Automation and Requirements Manager here at the FAA Command Center. Everyday we have over 50,000 individual flights that fly in the contiguous United States. At any one time you can have between 4 to 6,000 aircraft in the air.

A big part of what we do here is communication and that is some of the programs that we are developing such as a National Log Program here, are based on the idea that the 20 en route centers that control the traffic here in the United States and the smaller facilities underneath those centers are all communicating in real time with us.

The idea of this program is to be a national coordination point using an Oracle infrastructure at this facility, the Air Traffic Command Center.

In the past they've had tel-cons that are initiated from this facility, all these other facilities on-line and basically they're keeping paper notes or maybe they have their own little computer systems they've built in each facility.

We have thousands and thousands of pieces of information about opens, closures, any number of constraints around the system and that information has to be passed to us quickly and in a timely manner.

Every facility with the National Log Software can now report on the data that's in the database.

Well it's a way that our automation handles the communication between all of our hundreds and hundreds of airports we have around the system and here so we can keep up to the minute information.

This single system intends to continue to grow just to carry this facility and the rest of the FAA facilities that title this software far into the future.

(video ends)

CRAIG BARRETT: That's an application built actually off of Dell's servers. The last simple example I want to show you has to do with CERN, and you may be familiar with them, it's a physics operation in Europe, and they really explore what matter is made out of. And we're going to show you a little video of their facility in France, and you'll get an idea of the magnitude of this facility, and perhaps also when you're watching this, get an idea of the magnitude of the data that this facility is able to generate.

It basically can generate a petabyte per second of information that needs to be analyzed. That's basically ten to the 15th bytes per second.

What you'll hear a little bit later on is how they use filters and mathematical analyses to get that down to some manageable level of 100 megabytes per second of data, but very impressive application of this macroprocessing concept of modular building blocks to put enough capacity in place to do the work that they have.

Let's roll the video and then we'll talk a little bit with the project leader after we watch the video.

(Video playing.)

Bonjour, welcome to France. Today we have a backstage pass to CERN, the world's largest particle accelerator.

100 meters below this beautiful French countryside is a 27 km loop where particles are accelerated and slammed together at extreme speeds. Along this 27 km loop are 4 detectors that sense information about these collisions.

Here we are in pit five, this is where the compact muon solonoid is being assembled.

This is the CMS detector which is made up of several sub detectors. You can think of it like the layers of an onion. In here we have the inner detector. These pieces are the magnets and outside are the end caps.

This is the inner detector. In here is where the collision happens, and these are the sensors. And this is the world's largest magnet and these are the end-caps that go on either side.

You more or less try to reconstruct the same situation as the beginning of the Universe in order to find out how the particles are built, what are the actual fundamental particles and how they interact with each other.

Particles come in from both sides of the detector. They are collided there and after the collision they are sprang out and penetrating several pieces of the subdetectors; thus generating huge amounts of data.

The detector is generating 1 petabyte of data per second. Through a number of hardware and software filters, we reduce that down to100 megabytes of data per second. This data then travels 15 kilometers from here across the border to Switzerland.

Welcome to Zurich, this is a place where we are receiving all of the data from the detector and storing it. We would normally use a small number of big SMP machines. With a 9i rack technology of Linux Intel we can now think about using a large number of small nodes. And indeed we have been successfully with the 3 nodes Intel Linux cluster and are currently working on a 10 node cluster.

We are very excited about this technology and seeing how far it can go. Let's now have a look at the place where we can store all of this data.

This is the kind of machines where we are storing part of our data. It is very cost effective. It's a Linux Intel based machine. In one box 11/2 terabytes of data.

Well, that's it from CERN where the Oracle 9i on Intel really racks.

(video ends)

CRAIG BARRETT: What I'd like to do is have Jamie Shiers, who is the project manager at CERN to come up and chat about this application. Hi, how are you?

JAMIE SHIERS: Good afternoon.

CRAIG BARRETT: Maybe you could put in layman's terms what this project is all about.

JAMIE SHIERS: As we explained we have a very good model for how the universe works. One thing that's key is the understanding of mass and where that comes from. The reason this is important is without mass you wouldn't get clumping of matter into planets and galaxies so you couldn't have life and you couldn't even have OpenWorld.

CRAIG BARRETT: Wow. I'm not sure Larry knows about that.

(Laughter.)

CRAIG BARRETT: What sort of computational infrastructure are you trying to put in place for this project?

JAMIE SHIERS: After we reduced the data, we have come down to something about 100 petabytes. So to handle all this data we would need about 200,000 of today's Intel processors or maybe a few tens of thousands of tomorrows deployed in a worldwide grid.

CRAIG BARRETT: If you could get your order in by the end of the quarter, I would appreciate it.

(Laughter.)

CRAIG BARRETT: How far, realistically, along are you in putting the infrastructure in place?

JAMIE SHIERS: What we're trying to do now is work on the basic building blocks. So we have a test bed of about 1,000 modes, these IA-32, and they'll go to IA-64, and we have a 9i RAC that is now nine node, not ten as was mentioned in the video. And we're trying to use this to build up 100 terabyte databases which will be building blocks to build up the 100 terabyte system.

CRAIG BARRETT: I am intrigued by the fundamental work that you're doing, but each of the companies like Oracle and Intel have to report to the financial community. Do you have to do the same thing?

JAMIE SHIERS: We did with the physics processes, and a lot of people who work for CERN then go to financial institutes and simulate the market with some success.

CRAIG BARRETT: I'd rather have you keep them at CERN, but thanks a lot, Jamie.

JAMIE SHIERS: Thank you.

(Applause.)

CRAIG BARRETT: Obviously when you talk in front of an audience like this you have to talk about solutions, not just hardware. So it's much more than Intel microprocessors and someone's servers running Oracle9i RAC. It's really the total solution. So I think it's important to remember we're always talking about the basic products, the modular building blocks and the open architecture everything can innovate around. But then we have to talk about the solutions enabling and solutions channels that exist on that and that's really working with companies like Oracle and others to bring solutions into the marketplace, so working through various alliances, working on time-to-money solutions, working with our venture capital company to start up new companies to bring solutions to the mark.

It's more than just the hardware. It's absolutely the total solution that's important to bring it to the marketplace.

But the interesting thing about this whole macroprocessing concept is really from one approach, really from one hardware, one standards solution, if you will, you can get an unlimited number of solutions. And these are solutions that will run the gamut of what we've talked about today, working with Oracle on their messaging system. Working with Dell on their manufacturing system. Working with the FAA on their air traffic control system. Or working with CERN in the area of scientific analyses. Whether it be a financial problem, a manufacturing problem, an automotive problem, all of these sectors are covered.

We really think this is what the future of the data center is going to look like, this concept of macroprocessing.

If you look at what's happening to all of us as we go forward, there have been some basic driving forces out there, and I really think those driving forces are not going to stop.

The driving forces for the Internet, which is really the coalescence today of computing and communication, have been Moore's Law, which we're all familiar with. It started out by just doubling the number of transistors every 18 months, and 10 or 15 years ago that was slightly modified to doubling the processing power in a microprocessor every 12 to 18 months. We see how we can do that for another 20 years.

You see Metcalfe's Law where you see the value is equal to the square of the nodes. That's what the Internet is. We have many people on the Internet today. It's clear there will be a billion in a few years. So the value of the Internet is going to continue to grow.

The Photon Law, which is basically more aggressive than Moore's Law, says you can double the number of photons or double the bandwidth going down a glass fiber every nine or ten months. It doesn't show any sign of abatement.

All three of these things continue to push the Internet to be more important as we go forward. And that importance is really because the Internet is the fundamental driving force for the areas of communication, commerce, information access, and entertainment. And all four of those things benefit from more users and more bandwidth, more compute capability, more data storage.

So this is going to be I think a very synergistic relationship as we go forward. These three trends, not physical laws but observational laws, are going to continue to be valid, they'll continue to make the capability of the Internet greater for the things that it does well: Information access, communications, entertainment, commerce. And we'll continue to provide the underpinning technology to make that go ahead.

So this evolution of e-Business will continue. And you've already heard -- Chuck talked about it earlier this afternoon -- this concept of going from the old enterprise, glass-walled computer system. We've gone to the client/server phase. We're in really the server centric Web phase now. But going to Web services. And if you can picture yourself making this transition in a seamless fashion, legacy systems playing with current systems, common architecture, ease of implementation, this is really what the concept of macroprocessing is all about. This is what we're trying to bring to industry. The seamless integration of these capabilities and very simple scalability with the maximum unparalleled cost effectiveness of implementation.

So everywhere from the standard old database on up to the concept of Web services. And this Web services is obviously very important if you look at the Internet, and you look at the four things that I just mentioned the Internet is all about: Commerce, information access, communications, and entertainment. You want to go forward. You want to be able to combine all four of those in a seamless way, with a rich content basis, and that really requires more and more processing power, more and more reliable data storage capability. That's what really the Intel/Oracle combination I think brings in the marketplace.

What do we do from a standpoint of architecture? I think many of you are familiar with Intel at the desktop where our IA-32 family has been the primary processor architecture for the desktop computing.

We're also into handheld devices with something called the Intel Personal Internet Client Architecture. That's the high performance architecture that goes into personal digital assistance or cell phones.

We're also involved in Internet networking architecture, something we call the Internet Exchange Architecture, and that's really how you bring Ethernet capability across the board to the Internet. Everywhere from where we're familiar in our businesses from 10/100 up to a gigabit to 10 gigabit watching the ten gig coalesce with OSC 192, so we have one common standard at that 10 gigabit regime and moving on from there to 40 gigabit and beyond.

The thing we're most recently involved with is the high performance server architecture, and this is really the Itanium processor family or really the enterprise computing capability. We introduced our first 64-bit architecture member of the Itanium processor family earlier this year, but what I would like to do is really show what you the next generation is going to look like. What we have is a code name for our next-generation IA-64 processor which is McKinley, and I want to show you a couple of demonstrations of what that processor can do. We're just start to go sample it now. It will be in commercial production by the middle of next year, but it's alive, kicking, and running applications. It's binary compatible with the first generation of the Itanium family.

What I'm going to do is have Don Bowden come up and join me. We're going to do a couple of demonstrations with this. I know most are not very exciting but we tried to pick a couple that were interactive and you can see the real application. Don, why don't you fire that machine up and show the audience what we can do.

DON: All right. Well, Craig, I have to say I'm pretty excited to be demonstrating the next generation of macroprocessing. This is the first time anywhere we've had a four-node clustered machine based on the next-generation Itanium processor, code name McKinley, and it's running Oracle's 9i RAC.

One other thing I'm really excited about is the fact that this month we're going to be shipping McKinley to our customers for the pilot programs.

So what I want to do is take an opportunity to point out what we have here in our rack. Here is the start of our demo. We have our four McKinley systems that are making up our 9i cluster. Over here we have Pentium III-based Xeon server, application server, our Pentium III Xeon Web server, and here we have our EMC data storage rack.

CRAIG BARRETT: You have a whole data center here in other words.

DON: We do. It's up and running completely here. And this is all tied together with native InfiniBand architecture. Now, InfiniBand is an I/O infrastructure that simplifies and speeds up the connections between all your servers and your data storage. And that's all routed through our q-logic controller here, and we get a data rate of about 2.5 gigabits per second. So that's pretty fast.

CRAIG BARRETT: Probably not fast enough for the CERN application, but fast enough for our demonstrations today.

DON: Yeah, fast enough for what we're doing today which I'd like to show you here. What we have for an application today is a technology from a company called Bio-Key. Now, Bio Key makes finger ID technology, and it's used for the medical industry as well as a myriad of different IT security applications.

CRAIG BARRETT: Basically what you're telling us is you're going to use a fingerprint for an ID system.

DON: Finger ID for this. Exactly. The cool thing about this particular application, or one of the cool things -- it has many -- is the fact it was developed on an Itanium processor. We just brought it over, loaded it up on the McKinley and it ran. For the developers in the crowd, you don't have to recompile your applications.

CRAIG BARRETT: Let's show the audience what it can do.

DON: I get to be a doctor today.

CRAIG BARRETT: Your mother will be proud for about five minutes.

(Laughter.)

DON: What I want to do is go in and access my patient records, and from here I can select a patient. We'll take a look at Jane Smith. And go ahead and submit this. Here I'm going to use our finger ID tool, and this is going to recognize my finger. Yes, there we are.

CRAIG BARRETT: You're in.

DON: I'm in. That's right.

CRAIG BARRETT: Now you can access the patient's records?

DON: Yes, we can. And with that logging in, we had to search basically a huge database. I mean, the database that's going to store all this information. There's a lot of horsepower, a lot of processing power that's required in order to do that. So yeah, we can go in here and what I can do is select a treatment option.

CRAIG BARRETT: Looks like it's a broken ankle, so what are you going to do? A cast?

DON: I think a cast is in order here, so we'll select cast and now I'll go ahead and prescribe the treatment. As you can see, it wants to reauthenticate who I am.

CRAIG BARRETT: This means if you logged in, even if you had access to the patient's records it's not going to let you prescribe any treatment unless you reauthenticate yourself.

DON: That's correct.

CRAIG BARRETT: All right. Well, go ahead.

DON: All right.

CRAIG BARRETT: You're a success.

DON: That's all there is to it.

CRAIG BARRETT: That's one simple application. Again, big databases for all of that biometric application. I want to do another simple demo for you which is a little more in line with the opening skit we had today, music oriented. And again, something that takes a lot of database searching, a big database. David Sidd is going to come up, wherever you are. Oh, David, where are you?

DAVID: Sorry. Here I am.

CRAIG BARRETT: You forgot the wig.

DAVID: I was entranced by that demonstration.

CRAIG BARRETT: What are you going to show the audience?

DAVID: We're going to show you how a user can improve media rich data types out there. I want to start off by asking you a question. Have you ever found yourself in a situation where you heard a song on the radio, and you didn't get the artist's name or the title of the song. It's just in your head driving you crazy.

CRAIG BARRETT: Every once in awhile.

DAVID: What if I told you there's a company in Japan called Media Drive, and they have created an application that allows you to hum a few bars of the song, and return the music that you were looking for.

CRAIG BARRETT: All right. So all you're going to do is hum a few bars of any song, all the songs in the database, it's going to take that audio profile you put in, digitized, search that digital database and come up with the right song?

DAVID: Yes, it is.

CRAIG BARRETT: Right. Go ahead.

(Laughter.)

DAVID: Let me just ?? if you can give me some room, this is going to get ugly. What we're going to do is actually go to the application, I'm going to start up, it's called Cross Mediator, and I'm going to get it started here, and I'm going to enter my query. Ready for this?

CRAIG BARRETT: I'm ready.

DAVID: All right.

(Humming song.)

DAVID: Anybody know that song?

CRAIG BARRETT: Right.

(Laughter.)

CRAIG BARRETT: Maybe we'll get the other David out here and he can figure it out.

DAVID: Perfect.

CRAIG BARRETT: So you're searching now; right?

DAVID: Exactly. It took my humming pattern, did a spectrum analysis on it, submitted it to the database and that pattern is being compared against every characterized media file in my database.

CRAIG BARRETT: And I'm looking at the screen over there and it says Lenny Kravitz, right?

DAVID: Yeah, so it not only found the video but found the pattern within the video that may be possible matches and assigned an affinity rating for each pattern. As you can see, here's one that's a low pattern. Let's see if that's it. Well, that's the start of the video, but that's not exactly right. So let's go to a higher affinity one. Yeah, that's one. Here's another one. Here's another one. Yes.

CRAIG BARRETT: Yes, yes. All right. You're no Lenny Kravitz, though.

DAVID: Oh, come on. I strive to be.

CRAIG BARRETT: Do you think if I hummed in there "Only the Paranoid Survive," we'd get Andy Grove up on that thing?

DAVID: It's possible, it's possible. But this is a very sizable, sizable query that we're sending into our database. And this is comparing that pattern against every MPEG, MPEG2, MPEG3 we have in there. So you can imagine this is going to require a lot of horsepower.

CRAIG BARRETT: This is compute intensive and database intensive, right?

DAVID: Exactly.

CRAIG BARRETT: I'm coming to see you the next time I'm driving down the road, humming a tune and I can't figure it out, all right?

DAVID: All right. You know where to find me.

CRAIG BARRETT: Thanks, David.

(Applause.)

CRAIG BARRETT: Let me just simply summarize on what macroprocessing is again. It's really the design principle that we're going to go forward with in terms of really the ability of the entire industry to innovate around a common set of modular standards, really build a vast array of solutions on one set of common building blocks from one approach.

New products are going to be critical for this, new products like Oracle9i RAC, like our next-generation 64-bit processor families. And what our intent to do is to work with our friends at Oracle and others in the industry, basically to keep innovating to RAC your world, if I'm allowed to say that.

(Music playing.)

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