Blueprint for tomorrow

n the 1980s and 1990s, computer technology surged ahead in every area. Personal computers commandeered the desktop; Moore's Law correctly predicted that processing power would increase exponentially; storage capacities took quantum leaps; and prices plunged.

Controlling this sprawling mass of interconnected devices and structures were powerful new servers with the gusto of small mainframes. And gradually, unnoticed by the majority of users and even many IT professionals, these servers and clients were being knit into an interconnected network, the Internet.

By the year 2000, according to Larry DeBoever, director of enterprise architecture strategies at the Meta Group in Stamford, Conn., these disparate systems will interact in near-real time. Enterprise-computing managers will face what DeBoever calls "end-to-end velocity of information," and what IBM calls "pervasive computing."

Which technologies will answer the call for the increasing sophistication that will be required to manage this mass of connected data and devices? The answer may surprise IT professionals: The same ones that they are familiar with today -- only different.

"We are in no crisis of ideas, or power, or ways of getting there," says Jonathan Eunice, analyst and IT advisor at Illuminata, in Nashua, N.H.

Many of the elements of the future enterprise's IT infrastructure already are at our fingertips.

ASK THE EXPERTS. Expert systems have long been a lofty goal in computer science. These hardware and software systems are designed using highly systematic, rules-based algorithms to approximate stores of human knowledge and expertise in specialized areas. In the fields of medicine, chemistry, and engineering, expert systems are not uncommon even though the technology still is far from fully realized.

Expert systems are coming to the enterprise next. Larry Bielawski and Robert Lewand, professors of mathematics and computer science at Goucher College, in Baltimore, have written that expert systems (sometimes called knowledge management or diagnostic systems), combined with databases of ever-increasing sophistication, will form the underpinnings of the intelligent enterprise in the 21st century.

"There is a trend back toward data warehousing, with all resources available to anyone in a company who wants or needs them," says Randy Pack, chief engineer of information technology at RWD Technologies, a consulting company in Columbia, Md., that designs and implements expert systems. "It's an approach where companies make information resources available but add to [them] so [they are] more processed and useful to end-users. It is knowledge rather than just information."

For example, RWD has utilized expert system techniques in its development of Chrysler's Mopar Diagnostic System 2 (MDS2), an information and repair network used by service technicians. The MDS2 diagnosis-inference engine incorporates the knowledge of experienced technicians, which working technicians can consult before and during diagnosis and repair, and to which they can add information. In addition, MDS2 captures data from the sensors built into modern automobiles, and delivers that information via satellite and the Internet back to the central data repository at Chrysler's headquarters.

HAVE YOUR AGENT CALL MY AGENT. Agents and objects, carefully tailored for specialized tasks, complement and strengthen database technology and expert systems, and already are commonly employed in advanced database-management systems.

"We are finally starting to see, on the database and programming side, object technology coming along very nicely," Eunice says. "With objects, the advantage is that you are concentrating on the problem at hand rather than the implementation of your solution. Data is not stored in this cell or this variable, but in terms of the underlying problem."

Java and Jini, which promise a common computing interface linking mobile computers, home networks, embedded systems, and enterprises, are themselves object-based -- suggesting that object-oriented technologies will be component of the future.

Rudimentary agents, which automate the searches, transactions, and other mundane tasks, already are in use on the Internet. Expect agent technology to improve and become more ubiquitous over the next few years. IBM's Institute for Advanced Commerce, inaugurated earlier this year, says in one of its mission statements, "We anticipate a scenario in which billions of intelligent agents will roam the virtual world, handling all levels of simple to complex negotiations and transactions."

PARALLEL PROMISES. Parallel computing is another venerable theme in enterprise computing and, as with expert systems, people have criticized it for not living up to the hype. But many in the industry say parallelism will continue to wring substantial power out of enterprise hardware and software.

High-end servers and mainframes today typically employ parallel designs in any of several ways: in the execution of parallel instruction sets in a chip's internal processing logic; in software designs that utilize simultaneous operations and parallel search algorithms; in nonuniform memory architectures; and in massively parallel, "shared-nothing" memory designs; and in clustered computers. These designs may go far in providing the power that the next generation of globally connected systems will require.

"No doubt that parallelism is where the action is," says Randy Isaac, research vice president of science and technology at IBM's Thomas J. Watson Research Center, in Yorktown Heights, N.Y. "The trick will be to utilize massive parallelism at all levels, to understand the different kind of parallelisms and utilize them effectively."

MORE MOORE. Intel and Hewlett-Packard's Explicitly Parallel Instruction Set, the basis of Intel's next-generation IA64 chip (code-named Merced, and due in mid-2000), utilizes parallelism in its logic, compilers, and registers. In addition, IA64 may represent the future of the enterprise in another important way.

A vast network of commercial, governmental, educational, and personal computing devices demands a concomitantly vast source of processing power. In short, the enterprise is dependent on the continuation of Gordon Moore's prediction of exponential growth of processing power. Fortunately, many experts say they believe that Moore's Law will continue to hold well beyond the next generation of processors.

"Moore's Law is going to grind on for decades more," predicts Ralph Merkle, a research scientist at Xerox's Palo Alto Research Center, in Palo Alto, Calif.

A host of developing technologies will help ensure the march of Moore. These include cooling technologies, reduced operating-power requirements for chips, CPU-level multithreading, increased clock speeds, a move from bus-based to point-to-point topologies, the "doping" of silicon with other materials to enhance its conductive properties, and ultraviolet and X-ray lithography (used in silicon etching).

"There has not been this much attention [to] circuit optimization of microprocessors in 10 to 20 years," says Justin Rattner, director of Intel's server architecture laboratory, in Beaverton, Ore.

The steady increase in the number of integrated circuits at the chip level has been at the heart of the progress that Gordon Moore predicted.

In addition, the use of new computing paradigms, such as nanotechnology and the application of quantum mechanics to computing, will help keep Moore's law alive into the next millennium.

NEW ARCHITECTURES. The macroarchitecture of the enterprise is another area in which we already can see the clear outlines of the future. Now-familiar clustered systems will almost surely play an important role in powering the future enterprise.

Eunice and other analysts say clusters have excellent potential in marshaling computing power, exchanging data, and providing high-availability computing services.

Eunice claims that clusters, with their combined horsepower and increasingly efficient message-passing, may be the building blocks for the enterprise of the future. Clusters, ranging in size from campuswide to diameters as large as 200 miles, could, in turn, link to each other in networks of ever-increasing size: locally, regionally, nationally, globally.

Eunice points out that without such a modular approach, the latency of an interconnected long-distance network the size of the United States, for example, would cripple distributed processing power.

"Latency is such a huge and overriding issue within or between processors on a single system, it's 1,000 or 1 million times worse in long-distance computing," Eunice says.

As enterprise networks grow in scope and complexity, it's tempting to apply organic metaphors to them. The future will build on, many say thinking, learning networks that grow and adapt, much like the IT professionals who tend these computer systems.

"Conceive of the enterprise as an information ecology, an interdependent ecology of new machines," says Paul Saffo, futurist at The Institute for the Future, in Menlo Park, Calif. "Amidst all the change, the constant is the pendulum swinging back and forth. But every time the pendulum returns, it returns to a landscape that is more complex."

David Pendery (david_pendery@infoworld.com) is a reporter for InfoWorld who covers server operating environments and storage.