A network architect from Facebook said his expanding data center needs multiple 100 Gbit/second Ethernet links today. He called for starting a terabit Ethernet effort immediately, a suggestion quickly seconded by a senior technologist from Deutsche Telekom at this week's meeting hosted by the Ethernet Alliance.

Standards experts note it takes as much as a year just to gain the consensus needed to launch a new IEEE effort. The standards work itself can easily take another two years.

The IEEE 802.3ba standard for 40/100G Ethernet has not yet been finalized. A lot of pieces of that technology still require significant work, according to presentations at the Alliance meeting.

Nevertheless, engineers are keen to pave a path to the future. Chris Cole of Finisar suggested a step-wise evolution to 400 Gbits/s Ethernet as a practical step forward that would not require new and complex modulation schemes needed for a terabit spec. He even suggested the 400G module design based on 25G serial links below.

A proposed architecture for 400G Ethernet Click on image to enlarge.

"The value of a block diagram is it gets people thinking that we could do better," said Cole whose full presentation is included below.

The Finisar senior engineer was adamantly in favor of 25G serial links as the basis for next-generation Ethernet. "If you want to work on something where the money is going to be work on 25G not 20 or some other rate," Cole said.

It was clear from the meeting that several companies are working on 40/100G Ethernet products based on 25G serial interconnects, but some also expressed interest in 20G links. A representative of V.I. Systems (Berlin) discussed a prototype for a 40G VCSEL that could be pushed to rates as high as 100G.

The industry lacks the resources—especially given the global recession--to pursue multiple directions, said Steve Trowbridge, a senior technologist with Alcatel-Lucent. The same small group of industry experts will be needed to pound out a path to 400G or terabit Ethernet over 20 or 25G links.

So the question is open: Should terabit or 400G be the next target for Ethernet? Should it be created from 20, 25G or even higher speed serial interconnects? And when should this work begin?