IBM stumped for its approach to that future at a three-day technical retreat at its Almaden Research Center here this past week, picking up key support from Hewlett-Packard Co.

"For the last twenty years we have focused on performance," said David Patterson, computer science professor at the University of California, Berkeley, and one of the pioneers of the RISC microprocessor that ignited the drive to gigahertz computing. "We built things that went really fast, but they didn't work," he quipped in a presentation at the conference.

The problem is that increasingly complex networked computer systems have become more expensive to maintain than to purchase, said Vinod Khosla, a general partner with Kleiner, Perkins Caufield & Byers and a co-founder of Sun Microsystems Inc. Khosla, who also sits on the board of telecom carrier Qwest Communications, said Qwest estimates if it makes no changes to its infrastructure of 5,000 servers and 65,000 desktops, it will still pay $1.9 billion a year maintaining them.

Designing for change

"There's something wrong with that equation. We are not designing systems the right way," Khosla said. The answer, he said, is to design complex systems in a way that allows them to be easily inserted into an environment of continuous migration. "As engineers we were taught to optimize for performance and cost, but I am here to tell you the only thing that matters is optimizing for adaptability and evolvability," he said.

Taking one step in that direction, HP said it would contribute its research efforts to the Global Grid Forum, a group of researchers who are hammering out the so-called Open Grid Services Architecture, built on standards governing how diverse computers can automatically collaborate to share resources or run applications.

IBM intends to use the OGSA as a kind of middleware underpinning for an even broader initiative it calls Autonomic Computing. Its vision is to design computers that can configure, maintain and repair themselves without human intervention.

"Where we are going is toward a confluence of all these factors to build the distributed environment of the future. It's our belief grid computing [via OGSA] will become the basis of the evolution of Internet computing in the future," said Jeffrey Nick, an IBM fellow and director of advanced systems architecture, speaking at IBM's conference on autonomic computing.

Whether it's called autonomic computing, grid computing or Web services, an expanding pool of researchers see computer design taking shape around concepts in distributed computing that look more at automating the interplay between complex systems and less at bolstering the performance of an individual system or CPU.

"I think it's time to focus on a set of fundamental problems with how computers work," said John Hennessy, the president of Stanford University in a keynote speech kicking off the conference.

Hennessy, who helped define the era of RISC microprocessors, said robustness and scalability will be the top criteria for computer designers in the future, even as microprocessor designers find it increasingly difficult to wring new performance gains out of traditional techniques in instruction-level parallelism. "The flakiness we all accepted in the first-generation Internet will become unacceptable," he said.

"The design criteria are moving from price/performance to robustness and manageability," said Alan Ganek, a vice president of IBM's software group who spearheads its autonomic computing initiative.

Many researchers are pursuing the concept of grid computing as the next computing model beyond today's client/server systems. In this setup, systems in one data center might reach out to others across town or across the globe for help during peak hours. Thanks to a layer of software that turns physical systems into virtual resources that speak a standard language, they might do so with the same ease with which a PC processor taps its hard disk or RAM.

"We think of the data center as the computer of the future, and that platform needs the equivalent of an OS," such as a layer of interoperable software, said Rich Friedrich, a principal architect at Hewlett-Packard Laboratories in an interview before the conference. "We think this is where a lot of the action [in computing] is going to be in this decade."

HP rolled out its Utility Data Center software package last year, aiming to automate the process of distributing applications across a group of systems. On Tuesday (April 9), HP said it would extend that code to include the hosting of grid environments.

Commercial interest

To date, interest in grid computing has been largely confined to more than two dozen scientific trials, exploring, for example, sharing time on supercomputers for technical applications like climate control. But interest in grids is just starting to shift into the commercial realm.

"My feeling is there are a number of commercial test beds [for grid computing] but nothing that's ready for production use yet," said Ian Foster. Foster is a senior scientist at Argonne National Laboratory, a professor of computer science at the University of Chicago and one of the leading researchers in grid computing.

The OGSA effort — led by Foster and IBM's Jeff Nick — will release some sample code in May for a Java implementation of its core grid services specification. Researchers hope to have a final spec and an open-source reference implementation finished by the end of the year, said Foster. Meanwhile, the developers are hoping to get other top players such as Microsoft Corp. and Sun Microsystems to join the effort.

In an effort to rally such support, the OGSA has adopted some of the key underpinnings of Web services, including the Web Services Description Language.

Industry support is rapidly rallying around Web services as an important way to create modular applications that work together over the Internet.

As many as 67 companies are taking part in the Web-services architectural work group at the World Wide Web Consortium, reports Steve Vinoski, chief architect with middleware developer Iona Technologies and a member of that working group.

Busy intersection

"I personally have never seen that level of commitment to a standard before," Vinoski said. He said a roughly similar effort a decade ago under the Object Management Group attracted just ten companies. "Web services are at the intersection of autonomic computing and grid computing," he said.

The interest in these new concepts for automating distributed computing stems in large part from the pain end users are feeling pasting together increasingly complex networks of systems. Vinoski called the array of systems software being used for those purposes "applications integration spaghetti."

"You see astronomical figures for the numbers of ways these environments can be set up and you can't possibly test them all in any definitive way," added IBM's Ganek. "In just the database world there are hundreds of tuning parameters," he said.

That complexity has shifted the cost equation for data centers. In the early '90s hardware represented 80 percent of the cost of a data center, he said, but today half the costs of such a center are in the people who run it. In five years personnel costs could outstrip hardware costs by a factor of two or three, he predicted.

Daniel J. Clancy, chief of the Computational Sciences Division at NASA Ames Research Center, validated the problem of rising system-complexity issues. Forty percent of his group's software work is devoted to test, he said. "As the range of behavior of a system grows, the test problem grows exponentially," Clancy added.

Computer makers are seeing that problem as their ticket to reaping revenue in a rapidly maturing computer industry by offering services along with increasingly automated systems. "A lot of the grid interest is based on a services model. It's a very complex network environment and that's where I think a lot of the future revenue is tied," said Michael Krause, a technology manager in HP's server group.

Once standards are hammered out for Web services and grid computing — which may take months, if not years — industry agreement on common ways to handle so-called autonomic functions like self-configuring systems may be next.

"We have to get an industry coalition together of companies like IBM, Microsoft, Sun and others. Autonomic computing is not going to be as successful as it could be if there is no standard," said Steve White, who coordinates research on autonomic computing at IBM.

But IBM is still in the process of defining internally what aspects of its autonomic computing initiative might belong in the OGSA effort and which might ride above it, below it or parallel to it in the software stack, Ganek said.