Low-cost 10G Ethernet at risk

By JEFF CARUSO
Network World, 07/31/00

A low-cost option for 10-Gigabit Ethernet is meeting early resistance in the IEEE, potentially leaving users with only high-priced alternatives.

At issue are fiber-optic components designed for switches and other net gear that use 850-nm wavelengths of light and multimode fiber. In a meeting of the standards body's 10-Gigabit Ethernet task force earlier this month, enough vendors opposed this potentially low-cost, short-wavelength version to block its immediate approval. Instead, the vendors support longer-wavelength 1,310-nm optics for multimode fiber.

"The problem I see with the long wavelength is that it's not necessarily going to offer me, the customer, the lowest-cost solution," says Mike Bennett, network engineer for the Lawrence Berkeley National Laboratory and one of the users closely following 10-Gigabit Ethernet's development.

Products supporting 10-Gigabit Ethernet are expected to be used in LAN backbones as well as across metropolitan- and wide-area networks.

Copper-based Ethernet has usually been the cheaper way to go, but the high-bandwidth requirements of 10-Gigabit Ethernet make a copper version unfeasible. That leaves two major flavors of fiber optics: multimode and single-mode.

Single-mode fiber is often used for long distances, and the IEEE 802.3ae task force has agreed on two Physical Media Dependent (PMD) interfaces for single-mode. One uses 1,550-nm optics for distances up to 40 kilometers, and the other uses 1,310-nm optics for distances up to 10 kilometers.

Multimode fiber is used by companies and service providers within buildings or data centers. With 850-nm optics, multimode is often targeted at companies that don't need long distances and don't want to pay for the capability to transmit long distances.

But some vendors want to take the 1,310-nm optics generally reserved for single-mode fiber, apply them to multimode, and forget about 850-nm optics altogether.

While it is generally recognized that the long-wavelength optics would be more expensive, vendors aren't saying how much more they would cost. The task force's goal is to use technology that would cost three to five times as much as a Gigabit Ethernet connection (which today costs roughly $1,000 per fiber port). It is still unclear whether 1,310-nm could meet that goal.

An expensive network "gets away from what Ethernet is supposed to be all about," Bennett says.

But there are other issues as well, says Roy Bynum, a network architect for a major carrier, which he asked not to be identified. Key components of the 850-nm technology are not yet available for 1,310-nm on multimode, he says. Plus, 1,310-nm may require special patch cables to be installed.

The primary reason proponents cite the use of 1,310-nm optics on multimode fiber is that it would decrease the number of physical options customers have to choose from and therefore reduce confusion in the marketplace.

"There is a resistance to having every PMD that people want in," says Bob Grow, an engineering fellow at Intel and longtime participant in Ethernet standards processes. The task force has already weeded out many PMD proposals.

Building consensus

Much of the resistance is coming from vendors that are also members of an independent group called the 10 Gigabit Ethernet Alliance, which was formed to promote the emerging technology.

Its purpose is also to "build consensus," says the alliance's president, Tony Lee, who is director of product marketing at Extreme Networks. The alliance members try to work out their differences outside of the IEEE task force and then go to the task force's meetings with one goal in mind.

The group plans to hold meetings in August to debate the pros and cons of each approach to running 10-Gigabit Ethernet over multimode fiber, Lee says. This is in preparation for the task force's next meeting, to be held in September.

If the short-wavelength optics aren't approved for the standard, vendors may develop equipment that uses the technology anyway. However, such equipment would be proprietary, which is unacceptable for many users.

"Seldom, if ever, do we buy nonstandard equipment," Bennett says.

At this point, neither the 1,310-nm nor the 850-nm option for multimode have enough support to gain approval in the task force - the IEEE requires a three-fourths majority for approval.

"Nobody's ruled anything out yet," Grow says.

The task force hopes to reach consensus in September, to stay on track for a November draft of the standard.