It has beaten back a host of challengers in its more than two decades as a networking standard—to the point where it is the undisputed ruler of local, campus and metropolitan networks. Now the champion must transform itself to meet the futures demand for increased bandwidth.
Ethernet, the most venerable LAN standard and by far the most successful, will be getting yet another tenfold boost in capacity in mid-2002. But rapid adoption of 10 Gigabit Ethernet products is not a sure thing, and the regularity of future speed boosts is uncertain as well. The result: 10 Gigabit Ethernet could be more of a plateau than a steppingstone.
The reason for interest in keeping the Ethernet topology alive and growing is obvious from an engineering point of view: End-to-end Ethernet connections simplify network management and thereby reduce costs. Thats why it made sense to turbocharge the original Ethernet to create Fast Ethernet in the early 1990s, offering a boost in speed from 10M bps to 100M bps. Gigabit Ethernet products, which debuted in 1997, elevated Fast Ethernets throughput by another factor of 10.
10 Gigabit Ethernet shares the same name as well as the same frame format as its lower-speed brethren, but that is where the similarities end. The emerging specification, which is expected to be ratified by the Institute of Electrical and Electronics Engineers in the second quarter of 2002, was designed for carrier networks as well as enterprise LANs, costs 20 to 90 times more than current Gigabit Ethernet links and is expected to ramp up at a slower pace than its predecessors.
“The business drivers and market potential for 10G Ethernet look much different today than they did when we started working on the technology a few years ago,” said Chandra Kopparapu, director of product marketing at Foundry Networks Inc., in San Jose, Calif.
At that time, Ethernet was riding the Internet boom. As corporations chewed up network bandwidth, their desire for higher-speed connections was insatiable. Ethernet emerged as the best option because of its low price: Traditionally, vendors have been able to deliver a tenfold performance improvement at a price that was only three to four times as much as the previous technique. As a result, Ethernet emerged as the dominant networking option, first on the desktop, then in server farms with 100M-bps links and recently in enterprise backbone networks with 1G-bps connections.
With enterprises increasingly deploying Ethernet-only networks, vendors decided to push the technology out into the WAN with development of the 10G-bps specifications. The IEEE formally began working on the standard, dubbed 802.3ae, in the spring of 1999.
The desire to push Ethernet into the WAN meant that the standard had to support a broader array of connectivity options. Consequently, the standard comes in eight flavors, works with four types of transceivers and reaches distances from 300 meters to 30 km.
Standards participants are heading into the homestretch and finishing up the final draft of the specification. “There are no major technical issues that have to be addressed; the process revolves more around ensuring that the wording is correct,” said Mark Fishburn, vice president of technical strategy at Spirent Ltd., in London, a network test equipment supplier.
This year, physical-layer chip sets (transceivers), which send and receive data signals over LAN and WAN connections, became available from companies such as Broadcom Corp., in Sunnyvale, Calif.; Finisar Corp., also in Sunnyvale; and SwitchCore Group AB, in Lund, Sweden. Testing equipment designed to make sure there are no significant problems with these devices is emerging from companies such as Spirent.
With those building blocks in place, vendors such as Avaya Inc., of Basking Ridge, N.J.; Cisco Systems Inc., of San Jose; Extreme Networks Inc., of Sunnyvale; and Foundry announced plans for 10 Gigabit Ethernet products. And the suppliers have tried to alleviate potential interoperability problems by forming the 10 Gigabit Ethernet Alliance, in Mountain View, Calif. Founded in February 2000, the group, which now has more than 100 members, conducts interoperability demonstrations, including one at NetWorld+Interop in September.
Whats more, compliance-testing mechanisms from organizations such as the University of New Hampshire are starting to arrive, so users should be able connect different vendors products.
“I think users can feel fairly safe deploying 10G products in their networks now,” said David Passmore, an analyst at the Sterling, Va., office of The Burton Group Corp., a networking consulting company. In fact, a handful of next-generation carriers and a few large enterprises are already tinkering with 10 Gigabit Ethernet devices.
Users might feel safe with the technology, but what would they use it for? In the carrier market, startups and established service providers are building out new IP networks and using Ethernet to deliver services, such as MANs (metropolitan area networks). Demand for these services stems from enterprise deployments of bandwidth-intensive applications, such as video streaming and voice telephony, which need faster alternatives to low-bandwidth, dial-up and even dedicated T-1 (1.5M-bps) links. However, it has been difficult for carriers to deliver MAN services over SONET (Synchronous Optical Network) networks, which are inflexible. One subset of the 802.3ae specification offers service providers the option of running 10 Gigabit Ethernet over Dense Wavelength Division Multiplexing lines, a change that should make it simpler for them to deploy new high-speed services.
Yipes Communications Inc., in San Francisco, a next-generation network service provider, is interested in high-speed Ethernet technology. The company prefers using Ethernet rather than SONET multiplexers to build out its network and now offers customers bandwidth in increments from 1M bps to 1G bps. “By adopting 10G Ethernet, we would eliminate the conversions that now take place as we move data onto our backbone links, and that change would improve network performance,” said Kamran Sistanizadeh, chief technology officer at the company.
Another benefit of 10 Gigabit Ethernet is simpler network management. Network technicians can put their knowledge of managing Ethernet connections to use, employing one network management system to monitor a number of connections, rather than be forced to work with different systems for LAN connections, servers, backbones and WANs.
Because of the potential benefits, Yipes has been experimenting with 10 Gigabit Ethernet switches and is now examining how to integrate them into its network.
“For us, the major issues are not the switchs speeds and feeds but rather how the new system collects and passes information to our OSS [Operation Support System] applications,” Sistanizadeh said. In the enterprise, 10 Gigabit Ethernet is expected to edge its way initially into a few niche applications. One use is in a large campus network where a company lays its own fiber and uses high-speed links to connect office buildings—in effect, building a private MAN.
SANs (storage area networks) are another possibility. Traditionally, servers controlled storage devices, but this approach chewed up CPU processing cycles that could be focused on other activities, such as searching for information in a database management system. A SAN attaches to the back ends of a series of servers, usually high-end Unix systems or mainframes, and collects storage information. Placing that data in one location simplifies management and is more efficient than maintaining storage systems for each server. In addition, SAN connections support high-bandwidth links, so response is fast.
Some promising niches notwithstanding, there are plenty of hurdles in the way of 10 Gigabit Ethernet. Price is at the top of the list: Links are in the $20,000 to $90,000 per port range. “For the first time in its history, a new version of Ethernet does not support a compelling financial model,” said Seamus Crehan, an analyst with DellOro Group Inc., a Redwood City, Calif., market research company.
The prices are high because the underlying technology is extremely complex and difficult to produce. “With previous higher-speed versions of Ethernet, we were able to leverage existing technologies, such as Fibre Channel,” said Jean-Francois Thietard, product manager at Alcatel USA Inc., in Plano, Texas. “With 10G Ethernet, we had to develop a lot of the technology from scratch.”
In addition, existing networks pose challenges. “Corporations and service providers have deployed many different types of fiber,” said Foundrys Kopparapu. “Suppliers may claim that all have the same performance characteristics, but that is not really the case, so we had to build products that work over all strands of fiber.”
Still, service providers such as Yipes can build a sound business case for purchasing pricey 802.3ae links. “A carrier will spend $250,000 to $300,000 to install a 10G-bps link with SONET and optical equipment; a10G Ethernet option costs significantly less,” Alcatels Thietard said.
In contrast, enterprises have not been spending as much on their high-speed links. In fact, DellOro Group pegged the price for a Gigabit Ethernet port as deployed in many companies, at less than $1,000 as this year ends.
Given the economic climate, companies have been re-examining their IT investments. “Corporations are looking for an immediate return on any new IT investment; they dont want to wait two or three years for the payback,” said Todd Hansen, an analyst with Gartner Dataquest, in San Jose.
There are also questions about the need for higher-speed links in the enterprise. The Genome Sequencing Research Center at Washington University in St. Louis seems to be a typical 10 Gigabit Ethernet customer. About 400 employees use Sun Microsystems Inc. workstations and servers to conduct DNA testing, and the applications have usually pushed up against network bandwidth ceilings.
But this is not the case at the moment for the organization, which relies on Cisco Catalyst 6500 switches to support 16 Gigabit Ethernet links. “Right now, our network utilization is quite low, and the network has been able to handle any type of traffic burst—if we hit 25 percent usage on any of our Gigabit Ethernet links, that is quite unusual,” said Kelly Carpenter, system manager at the research center. “So I dont see us looking at 10G Ethernet technology for at least a couple of years.”
Other technology building blocks may fall into place during that time and spur demand for 10 Gigabit Ethernet links, however. For instance, current SAN technology (SCSI and Fibre Channel) cannot be easily integrated with IP networks and do not have sufficient throughput to fill up a 10G-bps line. But standards to address these limitations, such as iSCSI and Fibre Channel/IP, are now making their way from vendor development laboratories into deliverable products.
“To saturate a 10G connection, a company will need to install more powerful desktop and server systems,” said The Burton Groups Passmore. Just passing information from a computer onto a network is taxing: Moving 1G bps of data onto a LAN requires a microprocessor operating at a speed of at least 1GHz.
For these reasons, companies such as Alcatel are waiting until the 802.3ae standard nears ratification in the second quarter before announcing their 10 Gigabit Ethernet products. “We dont expect to see significant demand for 10G Ethernet products until 2003,” Alcatels Thietard said.
Analysts do expect increases in desktop, server and network power during the next few years. Recently, vendors began delivering network interface cards that support 10M-bps, 100M-bps and 1G-bps Ethernet transmissions to the desktop.
International Data Corp., in Framingham, Mass., expects the worldwide installed base of Gigabit Ethernet ports to increase more than fivefold by 2004. As companies install multiple Gigabit Ethernet links, they will look for a way to consolidate them, and 10 Gigabit Ethernet could be the answer.
“There is no doubt that service providers and enterprises will adopt 10G Ethernet, but they just may make the transition a bit slower than everyone expected a few years ago,” Passmore said.
Paul Korzeniowski, a free-lance writer in Sudbury, Mass., specializes in technology issues. He can be reached at [email protected]