Want CNET to notify you of price drops and the latest stories?

Intel expands networking choices

The chipmaker offers cash-strapped telecommunications companies three new programmable network chips aimed to ease equipment-replacement costs.

Michael Kanellos Staff Writer, CNET News.com
Michael Kanellos is editor at large at CNET News.com, where he covers hardware, research and development, start-ups and the tech industry overseas.
Michael Kanellos
3 min read
SAN FRANCISCO--Intel's message to the communications industry is fairly straightforward: The company's chips will be cheaper over the long haul, and that's good--because you're out of money.

The missive was delivered Tuesday by Sean Maloney, general manager of the company's communications group, at the Intel Developer Forum here. Maloney went on to unveil three new network processors--chips for directing traffic on the Internet--and discussed how Intel's overall networking strategy will effectively take much of the financial and technical risk out of the business.

Like IBM and others, Intel is promoting programmable network chips, which should help with the penny-pinching. Unlike in the past, when network engines had fixed personalities, programmable chips can run a variety of software. While the chips won't likely reduce the acquisition cost of routers and switches significantly, telecommunications companies that adopt them will be saved the cost of having to rip out equipment that's become obsolete.

The three new networking chips are based around the XScale processor, the successor to the StrongArm chip acquired from Digital Equipment in 1998. The basic architecture of the chips is based around intellectual property from England's ARM, but Intel inserts its own enhancements. XScale chips consume relatively little power but can run at 400MHz or 600MHz. Chips that eat less power run cooler--important in stacked networking equipment that must fit into dense environments.

Each of the three chips is optimized for a different function. The IXP 425, for instance, is designed to transport voice, video or data over DSL, cable and other PC connections and costs between $25 and $5. the IXP 2800 and IXP 2400, meanwhile, are designed for managing high-speed traffic in networking equipment. These chips sell for between $230 and $695. Another chip, the IOP321 I/O, helps manage traffic inside servers. That chip was released Monday.

To help gain momentum for the XScale processor, Intel is investing millions to assist software developers in bringing their products to the company's chips--a process that can take up to a year, according to software developers at the forum.

But despite the optimism surrounding the new products, in an interview before the keynote, Maloney said that the communications market remains in a slump.

"The industry is so sick," Maloney said. "It needs to cut costs enormously."

Last year, Intel's communications group alone had to eliminate "thousands" of positions in a companywide employee reduction totaling 7,000.

Many of the layoffs came because of redundancy created by the company's rapid-fire series of acquisitions. Since 1999, Intel has spent about $11 billion on 35 acquisitions, with many coming in the networking and communications field, according to company representatives.

"We have been tough in serious headcount reductions," Maloney said in the interview. The pace of acquisitions will likely slow to a crawl, he added. Acquisitions are difficult to pull off, and in many communications fields "we have all the skills we need," Maloney said.

Maloney was particularly proud that the XScale technology was delivered on time, in-line with the product roadmap unveiled 18 months ago.

Meanwhile, Thomas Franz, vice president of Intel's access and switching group, which launched the company's new XScale-based network processors, said the industry's emphasis on speed in packet processing meant that items like security screening were moving on-chip. Intel's NPU architecture supported this through "Hyper Task Chaining," working on multiple sets of packets at once.

The architecture combines features previously found on a variety of other designs--such as pipelining, SIMD and a cache subsystem taken from the Alpha processor--with innovations such as dynamic voltage and clock frequency adjustments that modify power consumption according to task. This is particularly important in network infrastructure, Maloney said, where increasingly high densities of components make heat an issue. SIMD, he said, is well-suited to packet processing and encryption throughput.

In the keynote, Maloney also touched on the company's wireless networking strategy, reiterating its support for the 5GHz 802.11a standard that stands to increase throughput to 54Mbps. He took a pragmatic approach, saying that Intel supported the development of products that included 802.11a functionality alongside the new standard to ease transition issues. The company, he said, would also continue to work closely with standards bodies to overcome legal and regulatory problems that are slowing down introduction of 802.11a in Europe.