Dell DirectConnect Conference

Craig Barrett
Austin, Texas, USA
August 26, 1999

ANNOUNCER: Ladies and gentlemen, please welcome Michael Dell. (Applause.)

MICHAEL DELL: Good morning, and welcome to everyone again to the second day of DirectConnect. We hope you're having a good time, and enjoying yourselves.

Intel is our most strategic partner when it comes to architecture. And we're very pleased to have with us today Dr. Craig Barrett, Intel's CEO. And I'd like to ask him to join us and give us some views on Intel's architecture, their technology, and their cooperation and partnership with Dell.

Craig. (Applause.)

CRAIG BARRETT: Thank you.

Good morning, it's a pleasure to be here today. I left Phoenix to come here and cool off here in Austin. (Laughter.)

CRAIG BARRETT: I want to talk today about building blocks for the Internet economy. And the choice for this topic is relatively simple. For years, the personal computer has been the driving force for our industry and personal productivity. But as we go look at the next five or ten years, I really think it's the Internet economy that is the particular driving force, and associated with that are new ways of doing business, new business models, new ways of interacting with customers. And this carries with it a different set of requirements from the architectural building blocks.

So the topic I'll talk today about is client/server networking, but really in the context of the Internet economy.

And the simplest way to approach this I think is to look at a simplistic vision of the future, and that vision is very simply that as we look forward, we can see a time, not in the far distant future, perhaps five years, where we have a billion connected computers around the world.

And today, there may be 150 or 200 million computers connected to the Internet, but adding them at the rate of a hundred million a year, it only takes five years or so to get to this magic billion number. And that's really, just is a phenomenal concept from a market concept of a billion customers, a billion people you can communicate with on onto instantaneous basis.

But more important than that is the amount of commerce we're looking forward to running over this array of network computers. And you don't have to look out five years to forecast a trillion dollars of commerce. You only have to look out about two or three years.

I'll give you a little bit of statistics about that in a moment but let's assume this is relatively accurate that within five years we'll have a billion connected computers and within the next couple of years just in the U.S. we'll see a trillion dollars worth of commerce run over this array of connected computers. And the backbone of all of this is obviously the Internet and the Internet transition.

My boss, Andy Grove, refers to these things as strategic inflection points, sea changes, whatever word you'd like to use to describe them.

This New Yorker cartoon which has got www.yourlocalstore.com I think is indicative of that. And there's lots of cartoons of this sort that represent this transition that's taking place.

I like this one particularly because of the two gentlemen who are sitting on the front porch. One happens to be the IT manager, the other the line of business manager for this operation. And it's those two people working in concert who have to make this transition possible and that will be a recurring theme in my presentation today, that it's really a combination of the IT infrastructure which enables the business to be agile, aggressive, and adapt to this new environment.

If you look at some of the forecasts for e-commerce or e-business, which is really a combination of both business to consumer and business to business, I think most of us recognize the business to business backbone of electronic commerce will by far be the dominant fraction of the total numbers that we look at. But the forecasts have a couple of interesting characteristics. One is that every year when we do a new forecast for the out years, the forecast kind of doubles from the previous forecast.

What I've shown you here are '97 and '98 forecasts and when the forecasts comes out at the end of '99 I expect the forecast for the year 2002 to be larger than the forecast we did at the end of '98. But it does show the forecast for the U.S. to be in the $900 billion range in 2002. You can put that in perspective. The United States U.S. gross domestic product is about $9 trillion, so that's close to ten percent of the U.S. GDP from essentially nowhere two to three years ago. So from a magnitude standpoint we're talking about a big, big change. And from a rate of growth standpoint, the exponential growth we're seeing here is also representative of a major training.

The rest of the world lags somewhat behind the U.S. in this area but they'll catch up rapidly. You can see the non-U.S. forecast in the same time frame is for two or 300 billion dollars worth of business, combined well over a trillion dollars by year 2002. This means we can't ignore it and we either have to take advantage of it or get steam rollered over it.

One way to look at that is today we talk about Internet companies, those companies who have their business model embedded in the Internet. I think if you look out five years from now, in the 2004 frame, you won't be talking specifically about Internet companies, because companies will all be Internet companies or they effectively won't be competitive and will cease to exist.

So we're all influenced by this sea change, and what we have to do is figure out how to accommodate it, how to change our businesses, how to adapt our infrastructure to take advantage of this.

And that's really what I want to talk about. And I really had three separate topics to address today. One is the concept of mass customization or mass personalization, which, in fact, the Internet and e-commerce, e-business allows you to do. It really allows you to treat every specific customer, and even more than that, every individual within one of your customers that you interact with as a peculiar, significant individual and customize the content that you provide them over the Internet, whether it's marketing information, availability, pricing, delivery information, product information, whatever it may be. So the mass personalization is one issue.

The concept of agility replacing efficiency and I don't want to get into a semantics issue here. Michael was talking to you yesterday about velocity and efficiency and I'm going to talk about agility. We're all talking, I think, about exactly the same thing. How do you position your company and your infrastructure to react almost instantaneously in the marketplace. How can you provide instantaneous information to your customers and give them a new customer experience. How can you stay ahead of the competition in this respect.

And this is more than just looking at cost of ownership of an IT infrastructure. This is looking at how rapidly you can adapt that infrastructure to meet your customers' needs and your business place needs. So we'll talk a little bit about agility, recognizing that efficiency or cost of ownership is a given in this space.

And the last thing that I'll talk about is the demand for compute power, and I'll talk not just compute power at the client but the server backbone, the server farms which are really the hosts for information on the Internet, the traffic directors for the Internet. We have an immense infrastructure to put in place if we're really going to have a billion connected computers and do trillions of dollars worth of commerce over that array of connected computers. And that will be compute power in the back room and the server space. But I think most of us are looking forward to a world where we have balanced compute power between the server and the client, and we'll try to talk a little bit about that and give you some demonstrations in that respect.

But let me start off by just giving you a thumbnail sketch of what e-business looks like today. And the most familiar model we have today is really what I would call vendor centric e-business. That is where a vendor puts up a home page, customers come to the vendor's home page for information, to place orders, et cetera. And this is if you're going to order a computer from Dell, this is what you do. If you're going to buy a book from Amazon or Barnes & Noble, this is what you do. You go to their home page, you take your shopping cart, you look at the catalogue of items and goods and combinations of services that they have. You decide what you want to buy, you click, and they send the product to you.

This is really a vendor-centric model. This is the way the Internet commerce has started. And it's very good for business to consumer where, in fact, you're a business and you have thousands or tens of thousands or millions of consumers who want to buy product from you.

I think the interesting thing, though, is that the bulk of the e-commerce in the future is not going to be business to consumer but probably business to business. And in that sense, the vendor model is really not what we want. We really want to see something which is much more of a consumer-centric model, where, in fact, the customer is the center of activity, and the vendors are constantly downloading information to that customer, and then the customer takes that information, uses their local compute power to analyze it, plan their business strategy, and carry out their day-to-day business.

And I can talk a lot about this, but let me try to show you an animation in this respect. Let me try to show you an animation. I can't read that. What's it say? Can't start it. Find the file back there in the room somewhere.

Well, I've even lost my mouse now. Use the backup mouse. This is not customer centric. (Laughter.)

CRAIG BARRETT: Just bear with me. Let's see what we can do. Ahh. Okay. That's vendor centric.

Let's just imagine your business, you're a floral wholesaler and you have suppliers around the world: Holland, Singapore, Australia, Madagascar, Brazil. And the way you do business today if you want those people to supply you product is you go to their home page, if they're on the Net like the Singapore supplier, you address them, their URL, you go to the home page, you pull that information back to your site, you download the information which is appropriate to your business manually, put it in your spreadsheet, then you go to your next supplier, maybe in Holland if you're going to buy tulips there or whatever, you download the information from that site, and you go and sample these different vendor sites to get that information. And we could continue to do this.

But what I want to do is relate that to a customer-centric model. Same floral wholesaler located in the U.S., same set of suppliers located around the world, but rather than have the customer go out and pulse the vendors, what we have in a continuous background basis is the vendors supplying information to the customer.

And the customer automatically accepts this information, has intelligent agents in the background that can sort through the information, put it in the spreadsheet automatically and conduct the business in a real-time fashion with a variety of vendors, the vendors continuously downloading their information.

Now, what this allows you to do, then, in a real-time basis is to have an inventory model of floral goods obviously have spoilage dates or shelf lifetimes and you can have that sort of information, but you can also combine this with your particular business model by having other information available to you and sort through it. For example, if you're in the floral wholesaling business, weather condition around the world which impact growth of flowers around the world could be important. You could have an intelligent agent which is sorting through background weather information. If you notice here in this first line, which is bear grass, which are suppliers in Australia, we have a weather advisory. If I just happen to click on that I find, wow, there's an anti-cyclone off the west coast of Australia. It's probably going to affect my bear grass supplier from Australia. I better look for an immediate alternate to that supplier to my wholesaling business.

So this is the type of simplistic model where we have a lot of background computing takes place, vendors downloading their information, combining that with other relevant information with intelligent agents sorting out from this mass of information on the Internet, and using that to, in fact, run our business.

So this is what I'm referring to as a customer-centric model rather than a vendor-centric model. Now, many people have taken the first step and gone to a vendor-centric model, and are doing quite well. I think Michael mentioned to you yesterday that Dell expects to do roughly half of their business by the end of this year over the Internet.

Intel started doing business with our customers in the middle of last year, and we went from basically zero business in the middle of last year to a billion dollar a month run rate via e-commerce by the end of the year. And we expect to do roughly half of our total business, about $15 billion worth, by the end of this year on e-commerce. And within one or two years, we expect that to be up to 90 percent of our business, not just half of it.

Why do we find this is efficient? Well all the little green boxes here show you, cost, efficiency, customer satisfaction, the possibility to personalize information to various of your customer representatives. And by that I mean you can personalize information for an engineering manager and a customer, personalize it for a purchasing manager and a customer, personalize it for a manufacturing manager and a customer. You don't just send out one blank page to the world; you personalize the message for each individual.

Today, what we have is rolled this out to 40-plus countries. By the end of the year, we'll have 300 or so of our customers accessing personalized information. And within those 300 or so customers, there will be 4,000 or so different individuals who will have personalized accounts, personalized home pages.

And the issue here, again, is not just providing sales or availability information, but it's providing marketing information as well. And what we want to do is just give you a quick demonstration of how this might work. I'm going to have my assistant Phu come out and we're going to do a quick demonstration of what the Intel e-business link or our e-commerce model looks like down here at the end of the row.

But I want you to recognize here that it's much more than just selling product. It can be as much as if you want to download detailed technical information, confidential technical information, to hundreds or thousands of engineering managers around the world, where before we used to have to go out individually with hard-cover documents and get their signature acknowledging receipt of that confidential information, today you can do that simply by just downloading it over the Internet. The fact that someone accepts it is their electronic signature that they've taken that marketing or architectural information.

Phu, describe to the audience what we have here.

PHU THAN: Okay. Here I have a home page after particular customer. On this home page you will find specific information and applications tailored to that customer based on their company and individual profile developed with Intel account manager.

Information such as marketing plan between Intel and the customers, sales, technical and promotional activities, information download from newsletters sent out weekly or monthly, and also tons of other information that they can select from and customize for their needs.

CRAIG BARRETT: So of the several thousand Intel customers, again, if I'm an engineering manager, I have a customized home page that I go to with my own password. If I'm a purchasing manager, again, a customized home page.

PHU THAN: And it will be different for the marketing manager, also.

CRAIG BARRETT: Why don't you show them what they can do in the way of buying product from Intel.

PHU THAN: What I'm going to do is take you quickly to order process. It's very simple. Typically you search for a customer part number. Just type in the part number. Search quickly. And because the system is connected instantly in real time to Intel manufacturing plant worldwide, you can go in and check for availability and also you can check for price, instant pricing, either today pricing or some pricing, projected pricing, in the future.

And once we've done that, we're going to fill in the order. And this is standard ordering procedure, familiar with many of you in the audience. We're going to bring up information to review to make sure we ordered the right part. And once we've done that, simply submit the order, receive a confirmation number. And if we want to, we can go ahead and track that delivery if we want to. And we simply just -- we'll find out when the part is delivered to --

CRAIG BARRETT: Okay. That's neat. That's very efficient operation, but it is standard transaction as we kind of know it today and computerized. I think you can probably do something more for the audience and where we're going to be in the future. I've been talking about this customer-centric approach. This is more of a vendor-centric model.

PHU THAN: Yeah. What I'm very excited about, Craig, is next we have a feature called supply line management where we help customers manage inventory more efficiently by connecting their inventory system to our ordering management system. So let me show you how it's done.

So let's say, for example, I'm going to search for a -- a customer search for inventory on a part, in this case a Pentium III Xeon processor, 550 MHz, our latest processor, and I'm going to bring up a graphical representation of that inventory quickly. And you can see now it's a secure Web site, available 24 hours a day. And you see the inventory level being scanned and returning the -- the information will be returned in real time.

So what you see here is the blue line is where we are today. The green bar indicates the current inventory level. The yellow bar indicates the future inventory level. And the system is very smart. It can alert us if a minimum inventory level is specified, and if we're short of that part as indicated by the flashing red dot on the blue bar, it asks us if we want to reorder the part. So in this case, of course we want to reorder it. Simply click yes. And the order is instantly confirmed by Intel. And you can see once we order, the inventory level rise -- in the future, rise up to meet the minimum requirement.

CRAIG BARRETT: It sounds like you've got another error bar here just like I had a minute ago.

PHU THAN: It's not really an error bar, Craig. The system is also very smart. It also knows that in order to -- in addition to CPU, I also need other part related to it in order to build my product. So it asks me if I want to go out and look for other parts. Of course I say yes. And voila. In an instant, I complete my order and transaction.

CRAIG BARRETT: Great. Thanks, Phu.

That simple model whereby what you can do is combine an inventory model with automatic ordering I think starts to give the customer the ability to run their business in a coherent fashion. And that's really the concept behind this customer-centric computing model, which I think the entirety of e-commerce will go to, especially the business to business side. And it's something that we're all going to have to deal with.

If you talk for a minute about this agility or efficiency issue that I mentioned earlier, there's a very simple model that I have to describe what we've been doing over the last few years. We've been putting infrastructure in place, client/server infrastructure, and we've been very worried about cost, security, the network security, keeping hackers out of our systems, low disruption, 24 by 7, four, five 9s reliability. I think that's preoccupied the industry in the era of where we've started to have electronic commerce become very important.

And efficiency is absolutely a requirement. You have to be efficient the way you do business. You have to have good cost of ownership. You have to use your resources the most efficient way possible.

But the model that we're going to is how do you use those resources more effectively. And that's where we're headed, I think, over the next three or four years as e-commerce becomes much, much more important.

And this is the issue of worrying about employee productivity, putting decision support capability in the hands of your line of business folks. It's being agile on your feet and being able to be competitive in the marketplace as this new form of commerce sweeps over the face of the world.

And the real issue here is getting your information technology people working with the line of business people. And to put an infrastructure in place as partners as opposed to a solid, stable infrastructure which is not tuned to the business requirements that we face going forward.

And what we're really trying to do here is to elevate information technology to be an absolute business partner going forward as opposed to just being an owner of the infrastructure that's being put in place.

So being agile as we go forward is terribly important. Being able to model and change the model of the business structure is important as we go forward.

And some of the things you find are significant here, then, are put in infrastructures which are capable of being changed rapidly, being expanded or upgraded rapidly, scaled to the level of business. Once you start to do Internet commerce and you start to put a special out or a new product, it's not unusual for you to have two or three or four times the amount of activity one day to the next. So I want a highly scalable infrastructure that you can put in place to accommodate this.

The other interesting issue is as you start to put your system available to the outside world for Internet commerce, and also enable your people to go to the outside world to conduct their business, your IT organization starts to lose control over the number of users of the infrastructure. As many outside people who want can come in to try to access information. Your own employees can go to the outside world and try to draw information back to run their business.

So flexibility and agility here are very, very important. And the usual issues are flexibility and agility is how do you combine these two and still get a low cost of ownership.

And if you look at what the industry has done over the last ten years, which has been a phenomenally interesting transformation, our information technology infrastructures have gone from vertical, stovepipe type arrangements to very horizontal, standard building block open interface type of infrastructure.

And this time sequence that I've shown in this supplied really shows what's happened to the computer industry during this period. And it's a very good thing, because that's the only way that we've been able to really adopt platforms which are highly scalable yet highly flexible, highly modifiable in a short period of time. And you can just, I think, all relive the experience where we went from stovepipe infrastructure to the first concepts of local area networks, and then network services and directories which allowed us to do e-mail and communicate with each other. And then put middleware in place, which allowed us to also interface with other applications, and application services which are, in fact, allowing our latest programs to interface with legacy databases.

But by putting this horizontal infrastructure in place, we've gone from a very fixed proprietary solution to an open solution that the entire industry can work on, multiple choices at each level. And more importantly what it allows us to do is the flexibility to take the ultimate applications that drive our businesses, whether it's an ERP application, an e-business application, whatever application it may be. We can modify those applications independent of one another without changing any of the underlying infrastructure.

For example, if I look at Intel today, with an SAP common database infrastructure for running our materials, finances, production, et cetera, I may only modify that once a year or so and upgrade it. But the e-business application that I mentioned earlier I may want to modify every two months.

I can do that on the common infrastructure here with this horizontal structure.

Now, this is kind of old hat to all of us. It's fortunate that it happened. It's a very economical, cost efficient system that the entire industry can bring innovation to.

There's another area of our business which I think is going to undergo the same transformation over the next couple of years, and that's as we look at voice and data converging.

Data is growing by hundreds of percent per year in terms of data trafficking. Voice grows five or six percent a year.

Data is essentially today the dominant information flow we have in our systems, yet we still have voice networks and data networks separate. And if you look at what the structure of these networks look like, typically they're proprietary. They look like the computer industry 10 or 15 years ago.

And I think what we're going to see if this space is exactly the same transformation we've seen in the computer industry, which is, in fact, voice and data will converge. There's no need to have two separate networks. You can carry voice over your data network. Them' converge, and they'll converge not with a vertical structure but with horizontal building blocks, with open interfaces.

So we'll have a unified network, standard building blocks and open interfaces, and then be able to do very simple application upgrades to that horizontal network going forward.

This is an exciting prospect if you're a supplier, obviously, in that space, because the communication equipment business is nearly as big as the computer equipment business.

I think it's exciting from a user standpoint because what you do is get to simplify your life going forward with a single network which the entire industry can work to provide technological advance, new capability in a cost effective fashion.

So I think the next decade is going to see just as much excitement in this space as we've seen in the computer equipment space over the last decade.

Now, let's look alt the third topic, which is the increasing demand for compute power. And I want to look here at two aspects, the client and the server. And we'll assume that we're going to have also increasing network capability to go with those two.

The concept, basic concept, is just that we'll have balanced capability here, balanced computing between the client and the server and we need to upgrade both of them as we go forward, and upgrade both of them gives us the ultimate flexibility or agility that we need to take care of new applications, to take care of new e-commerce opportunities for our businesses.

If you just look at some of the data behind this, if you will, everything from the size of decision support system databases to the number of users accessing these systems, these are forecasted to grow phenomenally over the next several years. Average DSS database to grow from a hundred gigabits up into the terabyte range. Number of users to grow from the few thousand to the hundred thousand range. And those may seem like big numbers to you if you're not involved in e-commerce in a big way today, but if you just look at people like Dell, people like Amazon, people who have millions of customers who are accessing their systems and then the personalization that goes behind this to be able to get a specific message to each one of those millions of customers, you can see how the databases can become very, very large very quickly, and the number of users as well can become very large.

This leads to the concept of the model of what you need to be both effective and efficient, to be agile and cost effective, is you need a scalable network, but on top of that you need to have scalable compute capacity, scalable compute capacity from the standpoint of being able to add servers that give you the backbone to do electronic commerce, scalable clients to take advantage of the added information that they get so that the local person can modify their business structure, modify what they're doing to be most efficient possible. Keeping in mind that you want to be efficient, you want to be cost effective, you want to be able to manage this whole system down the wire.

So this concept of flexibility, cost effective and scalable all have to be combined together.

And what I want to do here is just to show you another simple demo of how you can do this sort of thing. Chris is going to come out, and we're going to try to show you a typical user in e-commerce, what their structure looks like, how they can add capacity, what that means to them in a simple way in terms of scalable capacity. And why strong clients are important.

So, Chris, why don't you tell the audience what we have here. We have a big rack behind you for one thing.

CHRIS: Yes, I do. The scenario I've got set up today is actually typical of a small- to mid-size e-business networking environment, very typical of what a small line retailer would deploy. What I have on my right-hand side is a Dell rack model containing several four-way Dell servers and running 500 MHz Pentium Xeon processors.

These are also running multiple operating systems, Windows NT and Linux. Linux is used for our front end and Web server functions, and for our back-end processing we have NT running for that guy.

And over in front of you Craig is a high-end workstation client used to go ahead and manage these servers.

CRAIG BARRETT: By high end you mean it's not released to the public yet?

CHRIS: Not yet.

CRAIG BARRETT: But it does run well over 600 MHz?

CHRIS: Yes, it does.

CRAIG BARRETT: Okay. Good. (Laughter.)

CRAIG BARRETT: And it's very cool, too, isn't it? (Laughter.)

CHRIS: Okay. So what I've got actually running on this high-end workstation is an application from Computer Associates called unicenter TNG which is used to monitor and manage the NT systems I have over here.

CRAIG BARRETT: It shows you a schematic of the network you have over here.

CHRIS: Exactly. Here's a layout of my network. It's 3-D, very rich in graphics. As you can see, I'll move my mouse and I can point at my Linux server here and the NT server is sitting back here. It's got a red ball indicating there's a bit of a problem so let's fly into that machine and find out exactly what's going on there.

Okay. So actually, it not only allows me to detect problems but also identify what kind of components are in this system.

I've got several different processors in here, and I've got some external peripherals, and I can also fly in and actually find out exactly what kind of applications are loaded in my system here.

Using the unispace.

So it's got a red ball above the performance predictable management software, so I'm going to defer that discussion to a little bit later.

In the meantime, let's go ahead and take a look at some of the sample Web site which is placing such high demands for all this processing power.

Now, this is a Web site from Intel. It's called AnyPoint Home Network. This is very typical and becoming increasingly common for small online retailers.

So as you notice here, we've got some really rich graphics, very interactive so the user can get a much better experience before making their purchase decision. Okay. I'm going to minimize this guy and actually go back to my TNG environment.

And let's go ahead and now take a look over here at, if you don't mind stepping over here, Craig, we'll take a look at four-way server that's running.

What I have right here, and it's going to be on your left-hand side, is NT's task manager. And here is where I'm able to track the processor load, the usage and things like that.

And what I've got running actually is -- It's running a little bit slower today.

Supposedly I have a hundred users simulated running on this Web server. And actually, something is happening in the time I left. The demo was preset so obviously something has happened in that time since it's been preset. But anyway, it's not really showing what it should be showing. It should be showing about a 90 percent processor load, and serving about six pages per second or equivalent to about 360 transactions per minute.

And right now -- so that would actually indicate that the processor load, it's pretty heavy and it's doing quite well on this machine.

CRAIG BARRETT: I think what you wanted to demo to the audience, Chris, if I could get this straight, was you're running a four-way system. You might have a few hundred users on this. They might be running close to the capacity, 80 or 90 percent capacity, of the server to give up the right number of pages per minute -- pages per second, excuse me. And if you wanted, then, to add additional server capability, all you'd need to do is to go to Dell, get another standard four-way box, plug it into your rack mount here, hook it into the system. What that would do instantaneously is give you more headroom.

CHRIS: That's right.

CRAIG BARRETT: Increase the transaction count that you could do and allow more servers to come in place. For example, if we were looking at the AnyPoint model, the home networking model, if Intel did a special on that, we expected to have two or three times as many customers come in to Intel.com to look at that, you could just add more capacity by putting in these standard building blocks.

CHRIS: Exactly. Why don't we go ahead now and just step over to -- back in front of this other machine here. And as you can see, what's going on also is as that point I made earlier about deferring back to this discussion about Computer Associates TNG manageability software, predictable manageability, it's issued a warning saying there's a high processor load and high memory swapping and high memory usage. So that's really a clue to you, to the IT manager, that hey, I'm saturating the server, and it might cause some down time.

So it's really important with the increasing complexity in business applications, and we've seen scenarios where EBay and things like that have gone down.

So I'm going to now go over to -- actually, I'm going to wait for you.

CRAIG BARRETT: You're going to wait for me?

CHRIS: So what does that all mean, basically? Now that we have -- what can I do, actually? Where can I go from this four-way server, actually?

CRAIG BARRETT: Yeah, where can you go from this four-way server? (Laughter.)

CHRIS: Well, what I'll do now is I'll toggle my console here and I've now got my eight-way server up and running and you'll see that on your right-hand side. And that is supposedly running the same workload as what's running on this four-way server here.

What you're going to be noticing on the green bar is we're only saturating our eight-way server about 60 percent of the time, running hundred users, doing various activities such as transactions, browsing, downloading, purchasing.

And we're only taking up about -- and we're still serving around 6.5 pages per second, about 360 Web transactions per minute, which should be close to what we should have been seeing on the four-way server.

But let's go ahead, we have that extra 25 percent headroom so let's find out what we can do with that extra headroom and I'll have my colleague swat off another hundred users and let's see what happens to the eight-way server at this point.

If you keep your eye on the average number in the middle you'll notice it will steadily increase. And it's climbing up now slowly as the hundred users are being added.

It's going to go ahead and cap off at about 9.5 representing about a 60 percent increase over this four-way server. And what you'll also notice and what's important to point out is our processor load is not really going to be affected too much. It's going to go up about another five or ten percent. But the important thing is we have all that extra headroom and performance to handle an additional set of users.

CRAIG BARRETT: Okay. So scalability is important. Standard building block important. And especially if you're into e-commerce where you have the possibility of dramatically different loads put on by the outside consumers coming to your Web site.

CHRIS: Exactly.

CRAIG BARRETT: Okay. Thanks, Chris.

CHRIS: Okay.

CRAIG BARRETT: If you look at the standard building blocks that are available today and the cost effectiveness of these, and this is a little bit of, obviously, a commercial for both Intel and Dell, what you see are, I think, some exciting things from the standpoint of if you measure the horsepower and if that's done by the TPMC, the market basket benchmark capability, you can see that a Pentium III Xeon system from Dell, four-way system, compared to the other architectures and the height of the bar is in fact the measure of the horsepower of the system, but perhaps more important than the horsepower of the system is the cost per unit horsepower, which are the red dots shown on that basis.

Again, a standard high volume building block. The same sort of economics that came into the PC infrastructure now going into the scalable server infrastructure, giving excellent performance and scalability and agility to the end user.

And there's obviously a lot more to come in this space as we go forward. Our road map shows Moore's law is alive and well. We expect to see a doubling of processing power every 18 months or so, if not more.

Actually, Moore's law seems to be speeding up a little bit these days. So if you extent the 32-bit architecture out to products beyond the Pentium III Xeon product line, we expect to see that same sort of performance growth over time. And middle of next year we expect to introduce our 64-bit architecture, the so-called Merced architecture, which then goes off on yet even a steeper performance growth over time with follow-on product families behind it.

The reason that a lot of this is important is, in fact, that as electronic commerce grows and becomes more significant, it's the server backbone or the server infrastructure that has to grow to accommodate this. So we not only need to put a billion clients out there but we need to put tens of millions of servers into the back rooms to support all of this usage.

And if you do a simplistic back-of-the- envelope calculation about what fraction of the server capacity is in place today compared to what we will need in a few years' time, and you'll need more capacity in a few years' time just because you'll have more users, the users will be accessing richer information, different services, people will want some degree of headroom to take care of users, clumps of users coming online. If you do that sort of calculation, you can say over the next five years we have only roughly five percent of the server backbone in place today that we'll need five years from now. So we have 95 percent of that demand to put in place in that time frame.

So there's a huge demand for server capacity that's going to occur over the next few years, just to accommodate this sort of rich Internet commerce that we're talking about.

And that's really what the IA-64 family is designed to do in a variety of fashions. First of all, it's a 64-bit machine. It gives you addressability for bigger databases. It has some speculative execution, predication. In fact, this is wonderful for encryption and security aspects. And it has superior floating point performance which is great for data visualization, data mining exercises.

So we're really looking at this processor to be one of the main engines of e-commerce going forward. The product is taped out, we're going to give an update out at the Intel Developer's Forum next week in terms of where we are and debugging the silicon.

If you just look at where we are from a resource standpoint and industry support standpoint, we already have eight different OS's booting up on the Merced processor, and a number of people supporting this. It's a pretty bold statement to say it will be the engine for e-business, but I think from a server and backbone standpoint it will bring the economics we see at the PC level into the high performance server space for big databases and for rich data exploration, data access, data visualization in the backbone.

Just to give you an idea of the people who are involved in this program today in terms of the OEMs who are listed at your left in the blue block and then to the OS vendors, the IHV vendors, the workstation vendors, the tool vendors and then the ISV vendors who are over on the far right, I think it's an industry's who's who list who is in fact involved in promoting this architecture with its introduction slated for about a year from today.

The other comment I would like to make is we're excited about the venture capital fund that we've established to port and tune software into this architecture, $250 million venture fund in fact made available to ISV's and others who will support this capability.

So very exciting prospect coming forward. We think it's going to carry with it lots of economics and hope it will really expand the electronic commerce business by providing a very efficient, capable infrastructure.

Let me step off of the server side of the business and look at the client side for a moment, though, and what we expect to see there. And one of the most frequent questions we're asked is why do we need more MHz? Why do we need more capability?

And I'd like to answer that question in the context of why you need performance headroom on the client side just as you need it on the server side. And answer it in the context of electronic commerce and this concept of customer-centric electronic commerce.

If you look at where we are today on this scale, where we're vendor centric with HTML and basically just viewing flat information files on the client with a very simple human interface, which is basically a keyboard or a mouse, to where we're going to go over the next several years and see what the impact on the client is going to be from a performance standpoint, as you go to a customer-centric XML richer database capability, agents who are able to search out information and put it in a useful form for you and start to get rich user interface capability, voice input. All of those issues start to either eat up MIPS in the background from a continuous computing or background computing standpoint or require MIPS in the foreground for a rich data user interface or a richer visualization interface.

If you then take this to the next extent going out a few more years and find you're going to have very sophisticated agents, because what use is this immense access to all the data and all the information in the world unless you can actively sort through that information in a tailored fashion to your particular business model, so you need very intelligent agents. You need a richer human interface concept. You need machines that are easier to use, mask the complexity behind the processing power.

So if you just look across the realm here in terms of base applications, user interface, background computing, intelligent agents, customer-centric e-commerce models, communications capability, communications capability where you use the processing power to do compression and decompression to make up for the limited bandwidth we still have for the home and small business, you can see that over a four-year period, we're probably going to it need a lot of headroom.

And I could go on and talk in kind of abstract terms about this and how you're going to use that processing power for a long time, but probably it's more useful to give you a demonstration of what the average e-commerce business user will look like say four years from today, in the year 2003.

And what we want to do is have Chris come back out on the stage and do a simple demonstration for you. And this is a jimmed up demo, there's no hiding behind that, but this is what you could imagine your workers to be doing in this electronic commerce business model in a few years.

So, Chris, why don't you tell the audience what we have here. First of all, I want to preempt you, this is the computer. I personally hate the lime green, but it's kind of an interesting form factor. It is the full computer here.

So let's imagine this is what the computer on your desk looks like, first of all, in the year 2003.

CHRIS: Okay. In order to give everyone out there a better understanding of the feature of e-business, what I have is a demonstration of what my business desktop is going to look like in the year 2003. And for the sake of this demo I'm going to be a customer service representative for a company called Ace Camera.

So let's get started. In order for me to first gain access to this computer I have to first pass a couple of security checks. The first one is going to be a fingerprint recognition. I've got this fingerprint scanner right here. I'm going to place my finger right there.

Okay. That seemed to work pretty good. After a proper fingerprint identification, I will confirm by voice, Chris Wren.

COMPUTER VOICE: Good morning, Chris.

CHRIS: Okay. It looks like it's let us in. Let's go ahead and find out what my schedule -- or my day might be looking like today.

Computer, show me new messages.

Show messages from Greg. Greg Holland.

GREG: Hi, Chris, it's Greg Holland. Hey, our meeting out at Smith's has been changed to 1:00. Hope this works for you. To get there, turn right at the exit, and Smith Cameras is on your right.

CHRIS: I'm going to want to take some of those directions with me on my trip so computer, convert voice to text, and update my calendar.

Also, computer, please contact Greg and let him know I'm going to see him at Smith Cameras at one o'clock. Confirm by voice.

CRAIG BARRETT: That's pretty cool. I like the user interface. You've got biometrics. You've got voice in, you've got voice to text translation, rich visualization. I like it so far. But you can do more than that for sure.

CHRIS: I can do more. Exactly. (Laughter.)

CHRIS: Okay. Let's go ahead and find out what we should work on next. Computer, what should I work on next? Oh, that's right, I've got a conference call at nine o'clock with Lei Wong of Camera Mart. She's in their customer service department.

Before that call, however, I wanted to do some background information on them, and actually look up their recent sales figures and find out how well they're doing. So computer, create a new sphere called comparative retail analysis. Comparative retail analysis for Camera Mart. Go.

So at this time my computer is actually going out on the Internet and actually searching for key relationships all relative to Camera Mart, looking through some internal and external data sources, trying to find text, trade information, sales, finances.

CRAIG BARRETT: I kind of like these soccer ball things you've got here. As I understand it, the most important information is at the center and then the peripheral information is around that, and you kind of have these intelligent agents in the background gathering all this data; right.

CHRIS: Exactly. And there's an entry right there for the sales history. Let's look at that. Computer, show me sales history for Camera Mart. It's not bad. CRAIG BARRETT: Dell does better. (Laughter.)

COMPUTER VOICE: It's time for your call with Lei Wong.

CHRIS: Okay. Let's go ahead and give Lei Wong a call.

LEI WONG: Hello, this is Lei Wong.

CHRIS: Hi, Lei. It's Chris Wren from Ace cameras. How are you doing?

LEI WONG: I'm good. Thanks for agreeing to this meeting today. I wanted to talk to you about our models for Ace. They're really lowest specially compared to some of the other vendors we use. Let me show you this margin chart from my mobile pad.

See, here's Ace. Is there anything you can do to help? I think so. Right now you're placing one large order for all of our products which we ship directly to your central warehouse. Did you know you could place orders by individual stores? This would allow you to basically redirect, redistribute your orders. LEI WONG: Wow, that would be great. How do we set it up?

CHRIS: Well, I need access to retail store codes, shipping location, addresses on your orders.

LEI WONG: I'm giving you access to that information now.

CHRIS: Okay. I see it coming.

LEI WONG: I'm going to try to convert one of these large orders now. Thanks, Chris. Oh, I need to go. I've got an appointment in a few minutes.

CHRIS: That's okay. I think we're finished right here. Thank you, Lei. Bye-bye.

LEI WONG: Bye-bye.

CHRIS: Okay. So right now what you're seeing on the screen is my 3D inventory application linking Camera Mart's ordering system via Internet with our ordering system so we can track their inventory. The green bar here going down represents their central warehouse inventory going down to one of their stores being represented by a bar in yellow.

CRAIG BARRETT: Thanks, Chris. I think, although that's a little bit jimmed up, obviously, there the issue is there's nothing in that demo that technically is impossible today. And what we really need to do is integrate that capability as we go forward. More processing power makes the human interface easier to do, the voice recognition easier to do, the communication easier to do.

But this is the sort of workstation and the sort of environment that the knowledge worker involved with e-commerce will be involved with in the future. And the industry collectively has to work together to bring this capability, whether it's the OS's of the application or the hardware or the architecture.

But this is where we're going. This is, I think, a very adequate demonstration of the sort of processing power requirements you'll have at the client level, as we demonstrated earlier, the processing power requirements you have at the server level to do this sort of commerce.

Let me just quickly summarize. E-business scale is growing dramatically. It is this concept that competition is only a mouse click away, I think is very accurate. People either take advantage of this competitive environment, be first movers and be successful, or they'll get steam rollered. This concept that every business will be an e-business in the future, in five years from today, I think is the battle cry we all need to follow.

Computing power requirements I think are going to continue to go up, both at the server backbone level and at the client level with a balanced form of computing. Very interesting issue of what's going to happen in the voice/data networking arena as that whole business area goes to a horizontal business, very much like the computer industry has gone to a horizontal business over the past decade.

We're excited about IA-64 family. Server capacity is going to be extremely critical in the future. If you believe the back-of-the-envelope calculation I showed with the fact that we only have five percent of that capacity in place today that we'll need in the next four or five years. We're going to need big, strong, cost effective engines to drive that. We think the IA-64 is a member of that family.

And this is probably the most exciting time, I think, not just in the computer industry but in the users of the computer industry as we move forward with this Internet concept, the strategic inflection point of e-commerce as it comes upon us. It's exciting because we all get to change the way that we do business. We get to customize our message to our individual customers as opposed to one loud message to all customers.

And even within a customer, you can customize the message to the engineering department, purchasing, manufacturing department, management, et cetera.

So it really does, truly give you the ability to act small if you're a big company or act big if you're a small company in terms of the competitive environment.

So I think it's going to be an immensely exciting time frame. I certainly don't expect the technology to slow down from a hardware and architecture standpoint. The next four or five generations are under design. We're confident of our ability to produce those. We're comfortable and confident in companies like Dell to bring the latest in technology to the marketplace and our other industry partners like Microsoft, Cisco and others to provide the networking and software capability to make you all haggard, stressed, but having a lot of fun in the process as the technology marches forward.

Thank you. (Applause.)

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