2001 Ad

Electronic News, Sept 10, 2001 by Gale Morrison

Chip vendors bring wholly new ideas to bear

With some communications equipment and chip company stock prices hovering in the single digits--when just a year ago some swung in the $100 and higher treetops--one might think the Networld+Interop show opening in Atlanta at the Georgia World Congress Center today would be a little less than fun.

But a funny thing happened when the bottom dropped out of the communications market. Engineering teams redoubled their efforts and took long, hard looks at what is working and what is not for networking gear makers, and a whole slew of more advanced communications semiconductors and systems for the coming 10Gbit/sec.-capable, converged network is about to hit the streets.

Industry executives knew two, three, even five years ago, that standardization and outsourcing had to come to networking chip and line-card engineering. Certainly the convention of $50 million in-house design programs and literally billion-dollar acquisitions for the next product line could not be sustained. So, semiconductor suppliers -- the stalwarts and the start-ups--dived in. What the line-up in Atlanta will show this week is how many of them found the waters to be different than they expected, and how many of them hauled out their masks, fins, oxygen tanks, wet suits and even shark cages in order to stay alive in the market.

One thing network IC teams have learned and will be showcasing at N+I is where and when the RISC architecture works for network processing.

Only when designers put their RISC processor-based supervisory ASICs in the path of higher and higher bandwidth from the fiber did they determine that a major partitioning had to occur, said Rick Kepple, vice president of strategic marketing at IDT Inc. IDT (nasdaq: IDTI) is very close to this market having sold more than 250,000 intellectual property (IP) coprocessors in the last 18 months, said Dave Cote, the company's vice president of worldwide marketing. Cisco Systems Inc. of San Jose is IDT's largest customer, he said.

Kepple also said network processing players are approaching a fourth generation of thinking on processor types.

"This argument (of the unsuitability of the RISC architecture for network processing) that people are raising now is about the data plane," he said. "The 64-bit MIPS RISC processor that acted as the primary data plane was among the first generation of RISC architectures ... that was doing both the operating system--the control plane -- and the data plane, or data path, functions.

"And the (speed) requirement kept going up and up, and the bandwidth from the equipment became much faster than a general-purpose microprocessor could handle. So, they've split out the processing into two things: the control plane processor and the data plane. You have this mondo processor to do the control plane and small NPUs doing the data plane.

"The most famous four-processor RISC machines came out of an internal Cisco program called Toaster," Kepple said. "But now all of these companies are into their third generation of architectures, and we're almost into the fourth."

Kepple emphasized that this process has been key to the growth of a market for standard communications products from a variety of semiconductor vendors.

"As you begin to be able to predict the performance that's required, then reducing that to silicon becomes easier to do," he said. "As the problem becomes more defined, it becomes easier to migrate this to silicon." IDT is going to roll out its IP co-processor road map in two weeks.

The System Thinkers

A major theme at N+I will be the momentum behind system-level approaches by those companies that only wish to supply the system builders. Network Elements Inc. of Beaverton, Ore., is bringing its Lithium 10Gbit/sec. networking module to Atlanta this week, and Claude Denton, vice president for product architecture there, said that those companies only thinking about their ICs for 10Gbit/sec. could be in for a world of hurt.

"I'd like to introduce the concept that chips for optical networking are obsolete," Denton said. "The level of expertise required, with the optics, the electronics, the software, the multiple protocols ... (networking now), is much more of a system-level problem."

Denton hit upon a theme many networking processing IC vendors have been striking this summer, and it's one that is sure to be heard in Atlanta. From all appearances, the paths of data communications equipment engineering and telecommunications equipment engineering have converged at 10Gbit/sec.

"The convergence of datacom and telecom is happening at 10G," Denton said. "The large enterprises have discovered they need 10Gig, and they are also finding that they need telco reliability. The issue is this: the current networking equipment design process fails at 10Gig," he said.

"The traditional way of building net-working systems has been very, very vertical," Denton said. "There has been an acceptance of merchant silicon in that role, yes, but... at 10G the component and board-level integration is just an incredibly difficult task.