SAN JOSE, CA-There was a feeling of deja vu at the “100Gb Ethernet: The Next Challenge for Communications Systems” forum held August 29 in San Jose by the Optoelectronics Industry Development Association (OIDA; Washington, DC). With 10 Gigabit per second (10 Gbps) Ethernet transceivers now installed into datacom applications, the industry is looking toward 100 Gbps-or something near to that data rate-for the next generation of backplanes and network backbones.
There is no dearth of applications: Lane Patterson from Equinix (Foster City, CA) described how that company uses Ethernet switches to connect routers within data centers for its rapidly growing traffic. Petar Pepeljugoski from IBM described how IBM would prefer to leverage Ethernet optics for backplanes for its supercomputers, but said that in many ways, “it’s already too late,” and it may have to use products standardized around Infiniband at 60 Gbps or 80 Gbps if there is no Ethernet standard soon.
There was even discussion of using 100 Gbps Ethernet in metro and long-haul networks, with Drew Perkins describing how Infinera (Sunnyvale, CA) already has a 100 Gbps solution in that space.
For their part, component suppliers suggested solutions that echoed the progression toward 10 Gbps Ethernet standards such as 10 x 10 Gbps parallel optics, coarse wavelength-division multiplexing (CWDM), and conventional serial solutions, as well as the use of duo-binary and other more complex coding schemes. The IEEE recently approved a study group to examine a 100 Gbps Ethernet standard. John Ambrosia of Force 10 Networks (San Jose, CA) estimated that it could take about three years to complete the standard.
How soon any product could be affordable is another question, however. There was talk that a 100 Gbps Ethernet transceiver would have to be near $5 per Gbps to be effective for use in server cluster switch networks. Even at $1000 per transceiver, however, and sales of 100,000 units per year several years from now, that only amounts to a $100 million business.
Who would pay for product development, if the opportunity is limited to that size? Drew Lanza of Morgenthaler Ventures warned against expecting venture investors to invest heavily. Likewise, the market for data links for backplanes in supercomputers is too small to support development for early adoption. The burden may fall on the systems vendors to develop expensive proprietary technologies to differentiate their system products. But system vendors have moved away from this kind of development. Cisco (San Jose, CA), for one, is better known for encouraging standards and then selecting its components from multiple suppliers. A likely scenario would be that parallel optical transceivers are used in the first products. These products, which are currently commercially available at 30 Gbps and 40 Gbps data rates, were also at the front line for introduction of 10 Gbps optics. Eventually, other approaches like LX4 (an Ethernet version that carries 4 wavelengths of light on one pair of fibers) can enter as intermediate-term solutions. At 100 Gbps, a serial solution may only be cost effective in long-haul applications, if that, and won’t appear until much further in the future, given the low volumes and high cost of indium phosphide (InP) electronics.
Director, Components Practice