Magazine

Innovation, Lego-Style


What's the use of innovating? Seemingly none, if you're in the computer business. It has been a joyless, nearly profitless existence of late. Since 2000, the server business has shrunk from $69 billion to $49 billion, says researcher IDC. The only company making enough to brag about is Dell Computer Corp. Analysts expect the Round Rock (Tex.) company to see this year's profits rise 22%, to $2.5 billion.

What really irks rivals is that Dell doesn't even share in the heavy research and development required to create new products. Instead, it rides gleefully on the coattails of industry leaders Intel Corp. and Microsoft Corp. Last year, Dell spent a wispy 1.3% of its $35 billion revenue on R&D. By comparison, Sun Microsystems Inc. spent 14.7% of its $12.5 billion in 2002 sales on R&D, yet racked up $587 million in losses trying to sell its powerful servers. "We've seen this movie before, and at the end of the day, Dell gets the girl," says Merrill Lynch & Co. analyst Steven Milunovich.

That doesn't mean innovation in the computer business has been left on the cutting-room floor. Engineers are steaming ahead with leading-edge designs as radical as the personal computer when it was introduced in 1977. By 2005, IBM and Sun expect to make such huge leaps in performance that they will break Moore's Law, stating that chip speeds double every 18 months. That's when Big Blue will unveil Blue Gene, a supercomputer faster than today's top 500 supercomputers combined. And Sun is working on so-called "throughput computers" that cram the equivalent of eight huge servers onto a single chip -- an advance that could boost the power of big Internet servers fifteenfold by mid-decade.

The face of computer innovation is changing, however. Today's engineers are creating cutting-edge machines using commodity, industry-standard building blocks. Call it innovation, Lego-style. Chips from Intel and Advanced Micro Devices (AMD) Inc. have cut the performance gap with what computer makers IBM and Sun use. Microsoft's Windows server program is now robust enough to do many of the tasks of pricey Unix systems. And the Linux operating system is free and particularly good at serving up Web pages.

Snap it all together, and computer makers have a running start at building state-of-the-art machines. That frees them to use their smarts to solve other problems -- from developing self-healing computers to packing more power into ever smaller devices. "The challenge we face is how to do extraordinary things with affordable components," says William R. Pulleybank, director of IBM's Deep Computing Institute. "We could not have afforded to [develop Blue Gene] otherwise."

IBM knows that today's technical breakthroughs lead to tomorrow's big-volume sellers. By letting advances in automated maintenance made by its 1,200-person mainframe R&D team filter down to its lower-end machines, IBM is gaining share in the key PC server and Unix segments -- more than six percentage points of share in the past two years.

Even Dell can't prosper for long without innovation. The company spends most of its R&D budget honing its superefficient manufacturing and distribution processes to make them better. But Dell also is working on computer designs that string together hordes of Intel chips to deliver supercomputer power. "I don't believe there's any individual in this industry, including Michael S. Dell, who believes that innovation is dead," says EMC (EMC) Corp. CEO Joseph Tucci. "I see a tremendous amount of room for innovation."

Top of the list? Technology that helps customers get the most out of what they already own. Most servers today run at just 30% of their capacity. Even mundane tasks, such as manually loading new software, can take days. "Today's computers aren't nearly good enough -- ours included," says Sun Chief Technology Officer Greg Papadopoulos. "They suffer from excessive cost, complexity, and power consumption." Below are some of the areas where computer makers are working to do better.

BLADE COMPUTERS While most computers based on Windows or Linux are priced to sell, there's one problem: Companies end up buying lots of them -- and spend loads of dough getting them to work together. For every $1 spent on new equipment, operating costs run $3, according to IDC. Imagine if all those servers could be shrunk down to the size of a legal pad and stacked inside one console like books on a bookshelf. That's the main idea behind blade servers. This lets companies fit far more computing into far less space, cutting real estate requirements as well as the monthly electricity bill. All told, blades can reduce operating costs by 25%, according to blade pioneer RLX Technologies.

Dell and IBM only recently began shipping blade computers, so sales are just ramping up. The market for blades is expected to grow from 150,000 units in 2003 to 1.9 million in 2007.

GRID COMPUTING Grid technology is designed to make the most of underused computing power. Companies are beginning to use grid software to plug into computing power on the Internet or private networks as easily as electricity can be drawn from the power grid. It's technology management on a grand scale. The software constantly monitors the network, searching for computers that could help run a job that may be overwhelming the company's own computers. It then slices up the job into small portions so all the machines can help.

Charles Schwab (SCH) & Co., for example, wanted its brokers to do complex portfolio-planning calculations for clients more quickly. When Executive Vice-President David Dibble began experimenting with ways to exploit the company's unused computing power, he expected grid software would significantly slash the four minutes it took to do that task. It did. When the application goes live by yearend, customers will get their answers in seconds. "The results were astounding," says Dibble, "and we didn't have to buy any more computers."

THROUGHPUT COMPUTING No computer maker has been more beaten down by the onslaught of inexpensive hardware than Sun. But the company is hardly giving up on innovation. It's working on a new class of computers based on a radical concept called "throughput computing." While most servers are designed to run a single task as fast as possible, Sun's idea is to develop computers that can run far more jobs at the same time. To achieve that, it intends to cram the guts of eight of today's servers into a single piece of silicon.

Throughput computing could be perfect for Internet servers that must handle thousands of requests quickly, from Google searches to music downloads to complex business transactions. "Sun could have a unique capability if this works," says Peter N. Glaskowsky, editor of the Microprocessor Report.

The first chips will be in the form of a blade server, due out in 2005. The chip may sound like science fiction, but consider this: Thanks in part to Moore's Law, it will be possible to get 500 million transistors on one sliver of silicon by 2005, and 1 billion by the end of the decade. That's a lot of real estate to try radical new approaches.

FLEXIBLE COMPUTERS Among the concepts made possible by Moore's Law is the idea of nearly reconfigurable chips. Today's jack-of-all-trades chips are designed to handle some tasks well, while doing passable work on other chores. These new chips -- and the computers they would power -- would be able to morph by the millisecond, depending on what the computer is being asked to do.

One such effort is a research project at the University of Texas, called the Polymorphous TRIPS architecture. Each instruction received from a program gives hints as to the nature of the task. The chips come pre-loaded with various plans, each designed for a particular type of job. It's as if you could have a dozen different floor plans for your house -- say, to take the ceiling off the den to create a patio in sunny weather.

How do the chips change? The chip is divvied up into 100 tiny panels, each of which can be programmed to take different forms, say either memory or networking circuits. The computer's software analyzes program instructions as they flow in, arranging the panels on each chip in the best way to handle the task at hand, like a never-ending game of electronic musical chairs. If a video clip is called up, for example, the memory circuits would be reprogrammed as networking circuits. That way, the clip would be sped on its way, rather than delayed sitting in memory.

Prototypes aren't due out until 2005. IBM, Intel, Sun, and Defense Advanced Research Projects Agency (DARPA) are investors in the project, but there's no certainty such chips will get to market. If it works, TRIPS chips could bring supercomputer-class power to everyday products, from game consoles to home medical imaging equipment.

As for talk about high tech being mature, TRIPS researcher Moore refuses to believe today's off-the-shelf servers are the final stop in the computer's evolution. "The economic reality of commoditization is undeniable, and if you don't recognize it, you get slaughtered," he says. But if there are no innovations, he points out, there will be nothing left to commoditize. At that point, the computer business would be a truly joyless and profitless place. By Peter Burrows in San Mateo, Calif.


Tim Cook's Reboot
LIMITED-TIME OFFER SUBSCRIBE NOW
 
blog comments powered by Disqus