Scaling Ethernet to new heights — as much as 100Gbps — and developing 10G Ethernet for copper wiring and internal system backplanes were among the chief issues at last month’s DesignCon conference in Santa Clara.
Customers and producers of Ethernet say the need for scaling the technology to 100Gbps and for a 100 Gigabit Ethernet standard is approaching faster than expected. The use of Ethernet inside computers and switch chassis, and 10G Ethernet over copper could result in performance gains and cost savings for customers of corporate LAN gear in the near future, industry insiders say.
“It’s time to start a higher-speed study group in the IEEE,” says Mike Bennett, senior network engineer at Lawrence Berkley National Lab (LBNL), who spoke at a 100G Ethernet panel at the event, which mostly drew members of the component and silicon design community.
Bennett said the bandwidth needs among his peers at other U.S. Department of Energy labs and among carrier networks will soon call for a 100Gbps standard. Current use of 10G Ethernet, and the need to aggregate 10G links are driving this requirement.
There is some support in the design community for Ethernet to follow the progression of SONET technology’s transmission speed, as 10G Ethernet development stayed close to SONETOC-192 specifications. OC-768, at around 40Gbps, is the rarified highest-level speed of SONET available. Bennett thinks the bandwidth needs of high-end users and carriers will exceed 40Gbps by the time such a standards effort gets into motion; moving to 100G is better planning for the future.
“We’ve always been an Ethernet shop,” Bennett said. “And our whole upgrade plan is based on scaling by factors of 10, so it would just be a natural progression to go to 100G Ethernet.”
LBNL uses several 10G links in its data centres to aggregate switch connections and links to supercomputer clusters, used for computer-intensive energy research.
“We’re getting to the point where 10G Ethernet is taking off, and at some point you’re going to need to aggregate those links,” Bennett said.
Ethernet also is catching on as a backplane technology for a variety of gear, including blade-server chassis, core LAN and metropolitan-area network Ethernet switches and routers, and broadband wireless and DSL access equipment.
The use of Ethernet in a device backplane involves the transmission of traffic in standard-sized Ethernet frames on a device backplane, which interconnects various modules and interface components of a machine. (In a blade server, the Ethernet backplane links blades with a shared interface; in switches, it’s the path traffic takes traveling among ports on different interface cards, or to a central packet processing module). In 2004, the 802.3ap Task Force was formed in the IEEE to help standardize this implementation.
According to Adam Healey, a member of the technical staff at chip-maker Agere Systems, development of physical interface standards for backplane Ethernet is getting closer. Speaking at DesignCon, Healey said the 802.3ap Task Force is completing the details of 1000BASE-KX and 10GBASE-KX4 and 10GBASE-KR; these are Gigabit Ethernet and two 10G Ethernet specifications, respectively, for running Ethernet a distance up to around 3 feet, using a four-lane printed copper interconnect.
According to the standards crafters, the use of Ethernet in backplanes would let component makers create products at lower cost, because there is no need to create a new technology for moving data inside of a box, when Ethernet is already an established standard for moving data outside a device over wires.
Standardizing Ethernet technology could pave the way for more interoperability among components, although observers say a level of component plug-and-play similar to that in the PC hardware market is unlikely. Official sanctioning of the 802.3ap standards by the IEEE is anticipated in about a year.