Network processors gain new ground
Hard hit by thedownturn in the telecom sector, network processorsare finding new hope in the fast-growing multiservice space
By Gerald Lazar -- EDN, November 1, 2004
|
An old joke about a couple eating in a restaurant actually says a lot about the network processor business:
"What a disappointment," says the man. "This restaurant has a great reputation. But the service is slow, the food tastes terrible and, to make matters worse, the cold food is hot and the hot food is cold!"
"I agree," says the woman. "And such small portions!"
So it has been with network processors. The devices—designed to process network packets at high speeds—have been a disappointment. Pummeled by the depression in the telecommunications sector, the technology was late to market, slow to be adopted and sometimes missed the benchmarks set by its manufacturers. The devices got a bad rap. Finally, though, network processors are gaining some traction in the market, providing some of the long-promised benefits to network switches, routers and other such devices.
Unfortunately, with a projected market size of about $157 million this year, according to research firm Communications Industry Researchers (CIR) of Charlottesville, Va., the market is one-quarter the size originally expected at this time.
Unique strengths
Network processors are relatively inexpensive and easily configured semiconductors designed for networking applications. "A network processor is a programmable device that is optimized for the task of processing packets or network traffic," says Bob Wheeler, senior analyst at the Linley Group, a technology analysis firm in Mountain View, Calif. The chips are designed to handle tasks from layer 2 and up of the open system interconnect (OSI) network model.
"Network processors resolved two contradicting trends," says Amir Ayar, vice president for business development of EZchip Technologies, a fabless network processor company in Yokneam, Israel. They include "the high-speed performance of ASICs and the high flexibility of general-purpose processors. The idea was to build a device that takes the best from both worlds."
"With general-purpose solutions, you're left with more hardware and software than you need," says Steve Armil, director of R&D for Paradyne, a North Largo, Fla., communications access equipment vendor and customer of network processor maker Wintegra of Austin, Texas. "And that ends up costing more." On the other hand, ASICs can do the job. But according to the industry group Network Processing Forum (NPF) of Freemont, Calif., ASICs are not upgradeable, take a long time to create and the designs can't be reused. What's more, they are only cost-justified in very high volumes.
Because network processors can look at data packets as they quickly pass by, they have little or no effect on the overall speed of the network. Plus, the chips are scalable—just add more cores for more speed. They're faster than traditional communications processors—an absolutely essential attribute in some markets—and therefore often play an important role in mid- to high-level networking equipment (see "Communications Processors Find Their Niche").
While network processors are programmable, many OEMs don't use that feature. "Vendors are trying to make it so they aren't quite so hard to program," says Linley's Wheeler. That's why robust software libraries are so important in the network processor market.
Time to grow?
CIR is predicting the network processor market will grow from this year's $157 million base to about $173 million in 2005, reaching $256 million by 2008. But that's significantly less than what the market was supposed to reach according to some early projections.
"We've gone through the downturn, now reality has set in," says Doug Davis, vice president of Intel's communications division in Santa Clara, Calif. "There has to be enough volume at a certain line rate to make the business attractive."
Wanted: New customers
Today's market leaders in the network processor space by sales include Intel, Applied Micro Circuits Corp. (AMCC), Freescale Semiconductor, Hifn, Agere Systems and Vitesse Semiconductor. More recent newcomers to the party include EZchip, Wintegra and Xelerated, among others. Many of these vendors appear to be focusing on the lower end of the market, where devices with speeds of 2.5 gigabits per second and below are competing against general-purpose microprocessors and ASICs as well as with communications processors selling for around $100 apiece.
"Very high speed is still needed," maintains NPF Chairman Misha Nossik. "But there's much more emphasis today on moderate speed and complex processing." One such area, scrubbing data packets for possible security threats, may turn out to be especially lucrative for network processors. Russell Dietz, vice president and chief technology officer of Internet security device provider Hifn, in Los Gatos, Calif., says that network processors are proving themselves in applications involving firewall and denial of service processing as well as in edge services such as private networking technology.
"It can be three years to start real revenue ramp up, and very few can sustain this."
—Amit Banerjee, director, AMC
Even with a dozen or so vendors, the market is probably still too crowded. "Further consolidation is going to happen," says Amit Banerjee, director of strategic marketing for San Diego-based AMCC. "It can be three years to start real revenue ramp up, and very few can sustain this."
Where the action is
Over the next few years, Intel expects that network processors will be used everywhere from small office/home office premises equipment all the way to core applications. "The fastest growth is in broadband access and wireless infrastructure," says Intel's Davis. But that may be because there's fast growth in this market overall.
Paradyne, for example makes data and voice communications access equipment for service providers and large businesses. According to Paradyne's Armil, his company adopted the WinPath network processor from Wintegra to provide a variety of services such as Internet protocol (IP) and asynchronous transfer mode (ATM) services. "They had a part that looked like it could fulfill a lot of our specific functionality," he says. "We talked to them early in their product development cycle, and they provide support during our product development cycle."
Network management may be a fruitful area for market expansion as well. San Jose, Calif.-based Maxxan Systems, for example, is integrating Intel's network processor into its switch and storage service product. The company switched network processors after the first generation, when it found that the processor was "too limited" in its capabilities. Two years ago, Maxxan recognized it needed a fast network processor to "run all those storage services we intended to do," says Ravi Chalaka, the company's vice president of marketing. It was quite a list, including—but not limited to—the Fibre Channel protocol, virtualization, replication, encryption and data migration.
Network processors also can play in the security space. According Peter Middleton, a semiconductor analyst at research and analysis firm Gartner in Stamford, Conn., a network processor's "forte is being able to address situations where multiple protocols have to be handled and translation has to occur." Hifn's Dietz suggests that the aggregation of security services, firewall and virtual private network (VPN), together with intrusion prevention, will create a fertile ground for network processors.
Concord, Mass.-based Xelerated is most bullish about metro Ethernet and the high-performance enterprise, says Gary Lidington, vice president of marketing at the company. "We believe that it plays toward our strength of efficiency." Bill Klein, network processor product manager of Agere Systems in Allentown, Penn., agrees that metropolitan area networks (MANs) are promising. "At the core, they can live with what they have, but at the edge, that's where we see growth," he says.
At the low end, network processors must still compete with communications processors. Some companies, such as Freescale Semiconductor, are betting that building up their communications processor lines to solve the majority of low-end network processing problems is the way to go. "People were voting with their dollars," says Bob Gohn, Freescale's director of marketing. Others, like Intel, are eyeing the same markets with cheaper and more powerful network processors.
Ultimately, the term network processor may die out. It has gotten more than its share of bad press, and the technology perhaps stumbled too much out of the starting gate. But the capabilities, whether embodied in network processors, service processors or some other, as yet unnamed device, will doubtless be of use to network device designers for the foreseeable future.
Is the network processing industry on the right path? Send your thoughts tofeedback@eb.reedbusiness.com .
Some network processor market players
| Vendor | Website | Focus | 2003 Revenues |
| Agere Systems | www.agere.com | • Lucent spinoff • Highly scaleable architecture | $1.8 billion |
| AMCC | www.amcc.com | • Among market leaders • Low- to mid-range provider | $131.2 million (fiscal 2004) |
| EZchip Technologies | www.ezchip.com | • Stakes out high end of the market | $1.8 million |
| Freescale Semiconductor | www.freescale.com | • Motorola spinoff • Discontinuing further development | $4.9 billion |
| Hifn | www.hifn.com | • Bought IBM's NP technology • Moving network processors into security space | $20.5 million (fiscal 2003) |
| Intel | www.intel.com | • Among market leaders • Offers highly programmable chips | $30.1 billion |
| Wintegra | www.wintegra.com | • Relative newcomer • Targets low-end access market | Privately held |
| Source: Electronic Business |
|||
Picking up the pieces
In the mid- to late 1990s, the communications industry was booming, and network processing made perfect sense. Traffic was doubling two or three times a year, and OEMs were hard pressed to keep up. ASICs took a long time to bring to market, and general-purpose processors were expensive.
"In early 2003, just about every major vendor was looking to divest itself."
—Russell Dietz, vice president, Hifn
The network processor as conceived back then was expected to be a billion-dollar business, and by some counts more than 50 companies jumped in to fill the breach with network processors. Then the bottom dropped out.
It wasn't the fault of network processors, particularly. Peter Middleton, a semiconductor analyst for Gartner says, "The overall communications market obviously went through the crash starting in late 2000. It still hasn't recovered to the extent that some of the enthusiasts would lead you to believe."
On the other hand, many vendors were late to market, and the first couple of generations didn't reach their advertised performance specifications. Projects were cancelled, design wins evaporated. Those who had predicted big things for network processors began back-peddling frantically.
When the Internet bubble burst, most vendors had nowhere to go. Between 2001 and 2004, IBM left the network processor business, selling parts of its Picoprocessor PowerNP product line to Hifn; Lucent Technologies (which had acquired an interest in network processing by buying Agera) spun out the technology as Agere Systems; and Vitesse Semiconductor announced that it would stop development of new network processors. Those stories were repeated dozens of times throughout the industry.
It wasn't all bad news, of course. Hifn, for one, was delighted with its IBM technology purchase. "We were very fortunate," says Russell Dietz, vice president and chief technology officer of the Internet security device provider. "In early 2003, just about every major vendor was looking to divest itself."
The Internet plunge might even have helped the nascent network processor business. Bill Klein, Agere's network processor product manager, says, "OEMs couldn't afford the high upfront cost of ASIC development."
Whether the technology was helped or hurt by the bubble and subsequent contraction, the players are in place for now. "There is no room in the near term for more startups," says Gartner's Middleton. Any company eyeing the network processor business now would have to "invest tens of millions in developing a new architecture. If you expected to turn a profit, you'd have to look for sales close to the size of the whole industry last year." —G.L .
