SAN JOSE, Calif. – A handful of companies are edging closer to silicon photonics, hoping to enable a future generation of 100 Gbit/s networks. That was just one of several opinions about the future of networking from veteran Silicon Valley entrepreneur Andreas Bechtolsheim in a keynote here.
Separately, the OpenFlow protocol is not likely to gain traction but the broader concept of software-defined networks will, he said in an impromptu session with two dozen engineers after his talk at the Linley Tech Processor Conference here.
[Get a 10% discount on ARM TechCon 2012 conference passes by using promo code EDIT. Click here to learn about the show and register.]
“Luxtera and a couple new companies are going after [silicon photonics], and think they have a path to market by 2014, enabling the first cost-effective 100G switches,” Bechtolsheim told EE Times after his talk. Such products are needed because “today the cost of the optics are far greater than the silicon,” at 100G, he said.
The rate of progress in networking has been relatively slow due to I/O bottlenecks such as limited pins on a chip package, he said. “It’s a nice improvement-- doubling every four years or so--but it’s not at the rate of Moore’s Law."
A handful of 100G interfaces are in the works, but only silicon photonics holds the promise of making 100G more cost effective than 10 and 40G nets, said Bechtolshiem. He predicted that the majority of servers will be on 10G Ethernet by about 2014 when Intel’s Haswell processors could start a shift to 40G nets.
Separately, Bechtolsheim predicted the OpenFlow protocol is too low level to have significant impact, but vendors will roll out their own APIs to simplify network configuration and management.