Flexible Ethernet 2.0 will fill in details of management channels the shims will use to communicate with MAC and PHY layers.
A new generation of Ethernet silicon that could help further spread use of the already widely adopted networking technology, especially in optical backbone networks, is in the works.
A preliminary spec for Flexible Ethernet at 100Gbits/second rates was formally set by the Optical Internetworking Forum earlier this year. The OIF will fill in that spec with more details and support for 25G, 200G and 400G rates in a 2.0 effort expected to start in November.
Multiple vendors demonstrated FPGA-based implementations of the technology at optical networking events earlier this year. Support in next-generation ASICs and DSPs is said to be in the works with first products rolling next year and broad availability expected in 2018.
The technology solves several problems. Users across the complex networking landscape are finding needs for specific data rates not currently supported in standards that can take years to define. More specifically, big data centre operators need flexibility to ensure they can fill up long-distance optical links, the most expensive part of their networks and the part that takes the longest to upgrade.
Figure 1: Verizon sketched out an early concept for Flex Ethernet. (Source: Xilinx)
In addition, large data centres need to upgrade data rates inside their networks faster than organisations such as the IEEE can roll out new standards. The IEEE will finish work next year on a 400G standard it started in 2014, but for years, the Web giants have called for terabit-class Ethernet links they may need within a couple years.
“I’m hoping [Flex Ethernet] becomes the de facto MAC for beyond 400G rates,” said Tad Hoffmeister, an optical networking architect at Google. “I expect a small rollout over the next couple years, but by 2018-19 with next generation DSPs and router silicon support it could really take off,” he said, speaking at an Ethernet Alliance meeting.
Several chip makers are starting to make commitments to support Flex Ethernet. Merchant suppliers of DSP specialised for optical networks include Acacia, Clariphy and NTT Electronics. Systems makers such as Cisco, Nokia and Huawei are also expected to support the capability in chips they design for their switches and routers.
The new chips are expected to provide a rich matrix of choices. They will support more baud rates, encoding schemes and levels of forward-error correction.
Flex Ethernet essentially creates a shim between the media access controller (MAC) and the physical layer (PHY or physical coding sublayer (PCS)). The shim will act as a dial, taking information about packets and data rates from the MAC and directing a PCS to recode them as needed.
Hoffmeister said the technique could expand the market for Ethernet switches and routers that connect distant data centres, an area mainly served by relatively dumb optical transport systems today. The relatively low-cost Ethernet systems could maximise use of expensive long-haul fibre links, he said.
Inside the data centre, Flex Ethernet could help deliver links that are faster than emerging 400G products that data centres will need in 2018-2019, said Tom Issenhuth, an optical specialist at Microsoft and an OIF board member.
Figure 2: Flex Ethernet can maximise use of expensive long distance optical links. (Source: Xilinx)
Between data centres, Issenhuth sees the need for links running at 300-600G, speeds not supported today. “If we need to max out a fibre, we want that flexibility to dial in whatever Ethernet MAC rate is appropriate,” he said.
The technology is also expected to be used inside smaller corporate data centres. Companies sometimes need to break large links into sub-rates or translate a 100G link made up of ten 10G channels into four 25G channels.
“Flex Ethernet is link aggregation done right,” said John D’Ambrosia, a Huawei engineer who chairs the Ethernet Alliance. “Carriers are starting to look at FlexE as a way to channelise and isolate traffic as needed,” for security or other applications, he added.
Indeed, a representative of Japan’s NTT said terabit links in 10-80km metro networks may be needed to support 5G services coming as soon as 2020. It could also be used to interconnect carrier and Internet data centres running DWDM traffic over single-mode fibres at distances up to 100km, an emerging area of collaboration, he said.
The Flex Ethernet work got its start in 2013 with discussions among potential users from Comcast, Google, Microsoft and others. Cisco made a proposal called FlexMAC that morphed into Flex Ethernet and was adopted by the OIF earlier this year.
Besides support for 200 and 400G rates, the 2.0 version expected next year will fill in details of management channels the shims will use to communicate with MAC and PHY layers. It will also add explicit support for 25G building blocks.
The technique is expected to reduce tension between chip and systems vendors and Web giants who are their customers and have been struggling for years to accelerate Ethernet road maps.