Group will define 100G backplanes, cables

Rick Merritt SAN JOSE, Calif. – A new group aims to hammer out by March 2014 standards for running 100 Gbit/second Ethernet signals over backplanes and copper cables. The IEEE P802.3bj task force will plow a path for the next generation of…

Rick Merritt

SAN JOSE, Calif. – A new group aims to hammer out by March 2014 standards for running 100 Gbit/second Ethernet signals over backplanes and copper cables. The IEEE P802.3bj task force will plow a path for the next generation of data center and carrier systems that already are being outfitted with 40 and 100G Ethernet line interfaces.

“The need for 100G backplanes is already here,” said John D’Ambrosia, chairman of the new group and a chief Ethernet evangelist in the CTO office at Dell. “People have products with 48 10G ports already at 4:1 over-subscription rates, and to get them running at line rate they need 100G backplanes, he said.

Indeed, at a formative meeting of the group a Broadcom engineering manager noted some line cards shipping today could fill a 1.76 Tbit/s backplane. Some systems have moiré than a dozen such cards, calling for backplane bandwidth of tens of Tbits/s.

“A lot of people trying to get 100G products up in port density and down in costs,” he said. “We are nowhere near 10x 10G prices yet,” he added.

The 802.3bj group will define physical layer designs for a one-meter backplane and a five-meter copper cable carrying four 25G parallel lanes. The group aims to have proposals in by May and have an initial vote on them as early as November 2012.

At these speeds, creating the write channels for such traffic is a big part of the group’s work. “The focus is not so much the chip people as the people designing the backplanes so they have an upgrade path,” D’Ambrosia said.

Rack-mounted servers will be one of the primary users for the cable spec as a way to link to top-of-rack switches.

Besides carefully defining backplane and cable channels, the group is expected to spend significant time defining forward error correction schemes and signaling approaches. D’Ambrosia said he expects a lively debate between proponents of non-return to zero signaling versus those advocating four-layer pulse amplitude modulation.

“Our objective is to do one PHY standard, but some people would like to see two,” he said.

In its formative meetings, the group attracted representatives from Broadcom, Cisco, IBM, Intel, Juniper Networks, LSI, QLogic and Texas Instruments among others.