Information Processing

IBM NEEDS A NEW NETWORK--BUT NOT TOO NEW

Mighty IBM may have its share of problems, but the one in data networking is one that most of its competitors would kill for. Fully 62% of the largest U.S. companies use IBM's Systems Network Architecture (SNA) as the primary method for tying their computers together, according to Business Research Group in Newton, Mass.

IBM says there are more than 50,000 data networks built on SNA. They run airline reservation systems, transfer funds, and coordinate vital corporate functions from inventory to payroll. Says one typical customer, Tom Garrity, director of centralized data communications at Litton Industries Inc.: "SNA is our bread and butter."

OLD NEWS. Here's the problem: The original SNA blueprint is showing its age after 18 years, despite many improvements along the way. IBM needs to bring its networking technology into the 1990s and prepare for the turn of the century, when networks will blast text, data, and video at rates of 10 billion bits per second (table). But there's a delicate balance to strike. While IBM must satisfy cutting-edge customers with new products for high-speed multimedia communications, it can't alter SNA so radically that existing networks are rendered obsolete.

The job of fitting new onto old gracefully falls to Ellen M. Hancock, vice-president and general manager of IBM Networking Systems. "There are trade-offs, clearly," says Hancock. "If you started from a totally clean slate, certainly you could get to market faster." Still, she's driving her $5 billion, 12,000-employee line of business to hurry new technology to market by speeding up software development, delegating authority to seven semi-independent businesses, and market-testing new technologies even before their roles in IBM's strategic plan have been fleshed out.

IBM can't afford to take its time. That's the mistake it made in the 1980s. With SNA, IBM dominated nets built around mainframes. But it underestimated the enormous demand for personal computers and for software that ties them together in local-area networks that work independently of mainframes. SNA was designed for a pre-PC era when mainframes ran "dumb" terminals that could do no processing on their own. IBM's late start in local networks of PCs led to other missed opportunities, among them the now-hot market for systems that interconnect local-area networks. Such companies as Cisco Systems, Wellfleet, and Cabletron Systems are growing quickly by selling hardware and software for large-scale PC internetworks that rival mainframe setups.

PEER PESSURE. To play catch-up and show customers where it's headed, IBM in March made what it called its most important networking announcement since 1974. Big Blue said it will take a much more aggressive approach to Advanced Peer-to-Peer Networking (APPN), an extension of SNA that reduces dependence on mainframes. New APPN software due out next year will make mainframes work as equals with other network devices--even PCs. That will make it far easier for PCs to get at mainframe data. And the company already is getting plaudits for its first router, a special computer that connects different brands of local-area networks. Called the 6611 Network Processor, it's slated for shipment in June.

The most important products are still percolating in the labs. IBM scientists and engineers are working on a device called Planet that can switch 10 million packets--or about 300,000 pages--of information per second. That's 150 times as fast as today's fastest router and gives IBM the hardware it needs to handle the high-quality video and graphics that will be commonplace on the networks of the future. But today's SNA can't take advantage of the full speed of Planet, so it could languish unless IBM raises the SNA speed limit--quickly.

This time around, IBM scientists promise not to let network architecture be an obstacle. So they are moving forward with Planet development even before network designers have detailed how every device will connect to it. Says Alan E. Baratz, director of high-performance computing and communications at IBM's multimedia unit in Hawthorne, N.Y., one of those pushing the pace: "We want to make sure that technologies that we understand better than anybody else get used in the marketplace."

The only way to do that is to get products out faster. "We realize that the time we can take to get to market is decreasing dramatically," says Hancock. She says IBM plans to try Planet out on its internal network by the end of the year, even though it's still a prototype.

`FREEZE TACTIC.' Skeptics still are waiting to see a new, fleet-footed IBM. They've heard pitches about new architecture before. "Unfortunately, they don't get paid per pound of architecture, or their stock would be going up, not down," says Robin Layland, engineering consultant in charge of network architecture at Travelers Corp. Litton's Garrity is equally dubious. "This APPN announcement is a freeze tactic to get you to not make a decision," he argues. "My biggest question is, are they really going to be able to deliver a usable product?"

They sure will, if Hancock has anything to say about it. As the world goes digital, boundaries between computers, communications, and media are blurring. Hancock wants to offer products for all three. Sure, those 50,000 SNA networks may be a lot of baggage to carry into the future. But they also give IBM some important advantages, notably detailed insight into how the world's largest corporations and organizations want to move information at gigabit speeds.

In the next century, such turbocharged rates will be routine. While IBM has to retool its network designs for that era, so do its competitors. And despite its recent setbacks, IBM still spends $6.6 billion annually for research, development, and engineering. Says Travelers' Layland: "It's like a roulette table: They can afford to place a bet on every number." The trick is persuading customers to wait while the wheel spins.RAISING THE DATA SPEED LIMITS

1957

FIRST AT&T MODEM

750 bits per second. For transmitting computer data over phone lines

1974

AT&T 209A MODEM

9,600 bits per second. The speed at which mainframes sent long-distance data in

original SNA

1985

IBM TOKEN RING

4 million bits per second. For local-area networks of personal computers

1990

FIBER DISTRIBUTED DATA INTERFACE

100 million bits per second. To connect local-area networks between offices

1998

SONET

622 million bits per second. For voice and data connections over public

fiber-optic networks

2000+

GIGABIT LOCAL NETWORKS

10 billion bits per second. For all kinds of information, including 3-D moving

images

DATA: IBM, BW

Peter Coy in Hawthorne, N.Y.