Son of Gigabit Ethernet

10-Gigabit Ethernet is coming soon to a LAN (and MAN) near you.

By Jeff Caruso
Network World, 08/09/99

Ethernet is like a popular movie - the string of sequels is inevitable, and each sequel tries to be bigger and better than the one before. But in the end, it's all the same movie. Get ready for Ethernet: Episode 4. After Ethernet, Fast Ethernet and the most-recently completed Gigabit Ethernet, the industry is already starting to look ahead to 10-Gigabit Ethernet.

The IEEE has formed a study group to determine the parameters - such as media, distance limitations and even exact speed - for a 10-Gigabit Ethernet standard.

Right now, the study group is debating the technical nitty-gritty, but it hopes to submit a project authorization request to the IEEE this fall so the development of a standard can begin in earnest.

Most enterprise network managers probably think such a high speed is excessive. But each Ethernet speed increase seems excessive when it first comes up. And those embracing Gigabit Ethernet today may have to consider taking the next step in a few years.

In a recent survey of 225 companies, Infonetics Research found that 40% plan to implement Gigabit Ethernet in their network backbones during the next 18 months, with 19% hooking in servers and 9% running the technology to the desktop.

"We even saw a lot of Gigabit Ethernet going into small companies, where they might not necessarily need the bandwidth now," says Mike McConnell, director of enterprise management and LAN programs at Infonetics in San Jose. "But the price points have been coming down, so users are doing some future-proofing." Once those links are in place, companies may find they need to aggregate them into 10-Gigabit lines.

The first customers will be research institutions, which are always pushing the envelope. "I'd like to start field tests of 10-Gigabit late next year," says Michael Bennett, network engineer at Lawrence Berkeley National Laboratory. Bennett presented a user perspective at the study group's last meeting in early June.

The lab needs as much bandwidth as it can get to run virtual reality applications and put together a next-generation supercomputing cluster.

Bennett says the lab is using Gigabit Ethernet today because the technology is less complex than ATM. He would like to see field testing of 10-Gigabit by late 2000.

This isn't your father's Ethernet

The real drivers for 10-Gigabit Ethernet are service providers, and this represents a fundamental shift for Ethernet. The technology has been primarily a LAN technology, but providers are latching onto Gigabit Ethernet as a less-expensive and simpler alternative to ATM in metropolitan-area networks (MAN).

The influence of WAN-focused companies in the early 10-Gigabit discussions is fueling a debate within the study group over what the technology's exact speed should be. While 10G bit/sec Ethernet is feasible, a speed of 9.584640G bit/sec would match the SONET standard of OC-192. SONET, a standard way of transmitting data in frames over WAN fiber-optic lines, is widely used by telecom companies.

If the speeds were the same, the equipment used to translate between SONET and Ethernet would be easier to use and less expensive, experts say. Plus, the whole network - LANs in different campuses and the network connecting them - could be managed as if it were all in one building.

Another open question is how 10-Gigabit would actually be transmitted along a fiber line. The two top answers are that vendors could use a single, high-powered laser, or four lower-quality lasers multiplexed onto the fiber, says Dan Dove, principal engineer of LAN physical-layer technologies at Hewlett-Packard.

HP demonstrated the latter method at the spring NetWorld+Interop '98 trade show, Dove says.

The demo used four channels, each carrying 2.5G bit/sec of data. The channels were put onto the same fiber using wave-division multiplexing, which uses light to transmit data.

The drawback to the four-laser approach is that HP hasn't yet proven it can make equipment that would be affordable enough for use in enterprise networks. HP hopes to bring the cost down by year-end, Dove says.

A single, more powerful laser might not be cost-effective, either - and it would require temperature controls that the lower-quality lasers don't need, he says.

Common ground

Some things have already been agreed upon, although it's still very early in the process. One important point that vendors overwhelmingly agree on is that 10-Gigabit will only run full duplex, not half duplex.

This means the new Ethernet will lose one of the fundamental properties of the original. It won't use the mechanism called carrier sense multiple access/ collision detect (CSMA/CD), which provides a way for multiple machines to share the same Ethernet segment.

That change won't make for a dramatic transition, however. Although Gigabit Ethernet can support half-duplex operation, virtually no network hardware uses that part of the standard, says Bob Grow, engineering fellow at Intel.

The IEEE study group has also penciled in the distances it will strive for when specifying the cabling for 10-Gigabit. These include 100 meters on installed multimode fiber and 300 meters on newer, enhanced multimode fiber.

For single-mode fiber, three distances are being considered. These reach 2 kilometers for campus networks, and 10 kilometers and 40 kilometers for MANs.

Future net

As for using copper wiring, that is something that might be considered in the future.

The study group realizes that putting Gigabit Ethernet onto Category-5 cables was a tight squeeze. For 10-Gigabit, the group is contemplating the proposed Category-6 cabling, but it would still take several Category-6 cables to carry 10 gigabits.

Once the details are ironed out, the next step is to submit a project authorization request to the IEEE, probably in November.

Standardization of Fast Ethernet and Gigabit Ethernet took just a couple of years, but the standards process could take a little longer than that for 10-Gigabit, Grow says.

So what's next? You guessed it.

"Before we're done with our work on 10-Gigabit, we'll have to start looking at 100-Gigabit Ethernet," Dove says.