FRAMINGHAM 24 FEBRUARY 2011 - Despite the fact that Facebook famously said that it needs Terabit Ethernet, the wait for the technology could last years, experts say.
The technical, standards and economic challenges to be overcome are many, researchers at the Ethernet Technology Summit said yesterday, but they all could be overcome with enough time and effort.
Standards work could take years, says John D'Ambrosia, a researcher at Force 10 who chairs the IEEE study group that looks at higher speed Ethernet technologies. "We need to start now if we want to get there by 2015," D'Ambrosia says. "But we have a couple of issues going on here now."
One of them is that the industry craves higher density, lower cost 100G Ethernet, which will need standards as well and compete for the standards-makers' time and potentially slowing down standards for faster technology.
Accommodating Terabit Ethernet would require upgrading PCI Express standards for computer expansion cards, says Shre Shah, a researcher with Xilinx. Work needs to be done to improve their speed and keep down latency if they are to support Terabit Ethernet. (Also read: Terabit Ethernet in sight.)
Figuring out how to transmit at terabit speed is daunting technologically. Integrated optics in silicon chips is a difficult challenge as well because it might require as many as 40 lasers at 40Gbps each. "That number of lasers is high especially if we want to integrate it" into a chip, says Arlon Martin, a researcher with Kotura, which makes silicon photonic devices.
He says by transmitting two bits per symbol using a technique called phase-shift keying, that could halve the number of lasers needed.
Vendors working on the problem would prefer to build on existing resources such as established silicon foundries and standard sized silicon wafers, says Eric Hall, vice president of business development for Aurrion, which develops silicon photonics. Bonding different materials needed to create lasers and waveguides is challenging, but progress is being made.
From an implementation standpoint, using 40 lasers would be very difficult, D'Ambrosia says. And power to run the gear would also be challenging, especially as higher speed technologies are supposed to promote overall power savings, he says.
Cost is a factor as well, with customers likely to want them as a low as possible, but that is uncertain because the need for the speed may be great enough that it would be worth it to spend more.
With 400Gbps Ethernet on the horizon, 1Tbps Ethernet might not be a big enough jump in performance to warrant the effort, says Chris Cole, director of engineering at Finisar, which makes optical components and subsystems.
Sign up for Computerworld eNewsletters.