SAN JOSE, Calif. " A startup's lab demo of 10-Gbit/second Ethernet over Category 5e cable is fueling a heated debate over whether copper cabling should be included in the emerging 10GBase-T standard. The demo comes as a handful of startups line up to define a next-generation physical-layer chip for ultrafast Ethernet over dominant, but aging, copper.
Electronics and cable companies have been locked in a standoff over whether Category 5e wiring, used in more than half of today's businesses, is up to the task of 10-Gbit/s data transfers. So far, the IEEE 802.3an group drafting the standard has left copper out, though no final decision has been made about including it.
SolarFlare Communications Inc. sent 10-Gbit/s signals 50 meters over Category 5e using a full-custom transceiver and an off-the-shelf 8-bit, gigasample analog-to-digital converter. The Irvine, Calif., company (www.solarflare.com) claims it will cover 100-meter lengths when its custom 9-bit A/D converter, which recently taped out, becomes available.
Founder and CTO George Zimmerman, who helped develop digital subscriber line technology at PairGain, was set to make the algorithms behind the demo public at the first official meeting of the 802.3an group last week in Orlando, Fla.
"We think if people know a Category 5e option is feasible, it will help them make a better decision," said vice president of marketing Ron Cates. "There has been a lot of skepticism about the feasibility."
"From a technology standpoint, what they showed us was pretty impressive. This could take 10G into the enterprise and lower-cost areas, where to date the optical alternative has been too expensive," said Michael Laudon, director of hardware engineering at Force10 Networks Inc., an Ethernet router/switch maker that was quick to support 10-Gbit optical links.
One analyst gave the demo a thumbs-down, however. "I look at this as a political move to influence the standards effort," said Bob Wheeler, who follows Gigabit Ethernet for The Linley Group (Mountain View, Calif.). "They don't even have intellectual property for an A/D yet. It's just a theoretical proof that shows if you throw enough gates at the problem, you can get a 10G signal down a CAT5e cable. What they are demonstrating is far from a real product."
Indeed, Cates suggested SolarFlare may not have a product until the 10GBase-T standard is complete in early 2006. That product would likely include a version of the new A/D integrated with the current 4 million-transistor transceiver at line widths of 90 nanometers or less, he said.
Cisco Systems Inc., Intel Corp. and a loose alliance of end users at U.S. national labs back a plan to put Category 5e cabling into the 10GBase-T standard. But cable makers oppose the move, saying the older cables aren't characterized for the requirements of 10-Gbit transmission.
Category 5e cables are specified for 100-MHz bandwidth, though some companies have shipped enhanced versions tested for up to 350 MHz. The 802.3an group is currently specifying a bandwidth of 650 MHz for 10GBase-T, though the number could drop to less than 500 MHz. SolarFlare and other startups said they can handle the job with 350 to 400 MHz of bandwidth. "It's not that the cable doesn't have better bandwidth; it's just that they haven't measured it. The idea that 10G requires CAT7 just isn't true," said SolarFlare's Cates.
The 802.3an group has set goals for specifying 10G over 55 to 100 meters of 200-MHz Category 6 or 100 meters of 600-MHz Category 7 cable. Some members say it would be possible to add an annex on Category 5e to a final version of the spec. In a January meeting, before the group gained its official status as an IEEE task force, 25 of 61 people in the .3an group said they would support such an annex and 11 people from four companies said they would work toward such a technology.
"The simple fact is, the majority of the cable at this lab is CAT5e," said Michael Bennett, network engineer for the Lawrence Berkeley Labs. "If it's capable of 10G and cost-effective, why wouldn't I want to support it?" Bennett got a group of seven other end users, including many of the national labs, to concur. "In 2006, when this standard is ready, half of the installed cable will still be 5e. I don't see how we can afford to ignore that."
"There are people with less than 50-meter lengths who will someday plug in 10G, but we don't think it's realistic to expect people to redesign their [Category 5e] floor plans to implement this technology," countered David Hess, vice president of standardization for Nexans, a leading cable maker based in Paris. "We know there are versions of CAT5e that are not as well-behaved at 450 MHz as they are at 250 MHz. You have a lot of variables in the manufacturing process that are not being as carefully controlled at 450 MHz."
Though the Category 5e issue flared up at standards meetings in January and last fall, it was expected to stay more on the back burner last week in Orlando, where engineers were to discuss knotty issues of channel models, latency requirements, coding schemes and crosstalk.
"We don't need all the answers on each technology decision, but we definitely need to converge on some decisions very soon if there is any hope of drafting [an initial] standard by the July 2004 [meeting]," according to the minutes of the January meeting.
For instance, so far the group has discussed two signaling approaches. SolarFlare proposed trellis-code modulation based on 10-way pulse-amplitude modulation (PAM) at 833 Msamples/s. And Intel proposed low-density parity checking using eight-way PAM at 1,000 Msamples/s.
Finding ways to characterize noise from unknown external sources " so-called alien crosstalk " was also expected to be a major focus of last week's meeting, according to Brad Booth, an Intel strategic-marketing manager who chairs 802.3an. "This is where it starts to get interesting," said Booth of the group's first meeting as an official task force.
"The biggest technical issue is in conforming to the emissions standards," said Laudon of Force10. "That was a big challenge with Gigabit [Ethernet] products and will be even more so at 10G rates with its higher power."
As the challenges grow, so does the list of those looking to define the next generation of Ethernet transceivers. History has shown new silicon players dominating each new Ethernet generation, said Cates of SolarFlare. National Semiconductor and AMD led the 10-Mbit generation. Level One dominated 100-Mbit Ethernet, and Broadcom and Marvell became leaders in Gigabit Ethernet, he said.
Teranetics (Santa Clara, Calif.) is one of those with its eye on the 10-Gbit prize. So far the company has kept a low profile, but it has heavy hitters behind it, including veteran venture capitalist Irwin Federman of US Venture Partners and John Cioffi, patent holder of key DSL technologies. Both sit on the Teranetics board.
The startup was founded in December 2002 by Sanjay Kasturia, a co-founder of 802.11 developer Airgo Networks, and others. The company shares with SolarFlare the goal of developing a 100-meter solution for Category 5e cable, said Rahul Chopra, co-founder and vice president of business development and operations.
Startup KeyEye Communications showed the first 50-meter, 10-Gbit Ethernet demo in December, though it used Category 6 cables.
The Sacramento, Calif., company was formed by a handful of former Level One managers including Robert Pepper, who was once that company's CEO and is now KeyEye's chairman. The startup hopes to apply its core technology to both backplane and cabled networking.
Other relatively young companies represented at 802.3an include transceiver maker Vativ Technologies, Plato Labs Inc. and iTerra Communications, which has produced a variety of components for mainly optical networks.
Analyst Wheeler said it's far too early to start evaluating the impact such companies could have. "There's not even a draft standard yet, and you have to remember companies like Broadcom and Marvell are going to take an active role, too," he said.