The chairman of the group that set 40 and 100 Gbit/s Ethernet standards gives his perspective on the recent decision to explore a 25G interface.
In 2007, the IEEE 802.3 Higher Speed Study Group was contemplating the next speed beyond 10 Gigabit Ethernet. It came to the realization that the bandwidth requirements of networking and computing applications were growing at different paces. The bandwidth requirements of networking applications were doubling approximately every 18 months, while the bandwidth capabilities of computing applications were doubling approximately every 24 months. This roadmap (below) has provided direction to the industry for the past several years.
Bandwidth projections in 2007 from the Higher Speed Study Group.
Last week, this roadmap evolved as the IEEE 802.3 Working Group approved the formation of the 25 Gbit/s Ethernet Study Group. This group will focus on 25GbE over a single lane for server interconnects. This effort is effectively rewriting the server roadmap developed by the task force I chaired that developed the 40 Gigabit Ethernet standard. This shift has pulled me into a number of conversations essentially revisiting the decisions made several years ago by that group.
I spoke on behalf of the 25 GbE call for interest (CFI) and was a supporter because the needs of the industry have changed since 2007, and Ethernet should evolve to meet those needs. The 25GbE specs will reduce the cost per bit from a capex and opex perspective for today's cloud-scale and tomorrow's enterprise data centers. A 25GbE server interconnect can reduce the number of top-of-rack switches by more than a factor of three, lowering power and cooling costs.
I want to clarify some misconceptions around the recent debates regarding 25GbE.
- Misconceptions Nos. 1 and 2: The IEEE voted down 25GbE at the March 2014 meeting in Beijing.
- Fact No. 1: No vote was ever taken, because no motion was ever made. People are referring to the results of a straw poll. Some will attempt to characterize this as semantics, but from a standards process perspective, it is very important.
- Fact No. 2: The IEEE does not vote. Decisions made in this group are a reflection of the greater industry at large, not the IEEE, which isn't making any decisions. Decisions are driven by industry consensus.
- Misconception No. 3: The IEEE started the 25GbE Study Group in response to the formation of the 25GbE Consortium.
- Fact No. 3: The lack of formation of a study group out of the March 2014 session led to further industry consensus building, as well as additional time for evaluation by the industry to assess the market needs and requirements for 25GbE. In addition, the request for a CFI requires a 35-day notice, so the request was made well in advance of the announcement of the new consortium.
It takes time to build consensus, and that should not be underestimated. Consider the decision to include the development of 40GbE with the development of 100GbE. This debate really began to ramp up in January 2007 and culminated with an agreement in the Higher Speed Study Group in July of that year. The 25GbE decision followed a similar pattern in which a January CFI was discussed in March and culminated with the formation of the 25GbE Study Group in July.
This new group is poised to move very fast. There has been significant work within the IEEE since 2010 on the development of 25 Gbit/s electrical signaling for chip-to-chip, chip-to-module, backplane, and copper twin-ax applications. Some technical matters need to be resolved, such as forward error correction, but the wealth of past IEEE work will provide a foundation for leverage.
The larger issue of Ethernet's roadmap needs to be considered. The current IEEE P802.3bs 400GbE Task Force I chair is considering the development of 50 Gbit/s electrical signaling for chip-to-chip and chip-to-module interfaces. Some are arguing that this will begin the next cycle, and the next server rate that makes sense will be 50GbE.
However, there is already a defined rate beyond 25GbE known as 40GbE. Some have tried to argue that 40GbE was a mistake, but the simple reality is that 40GbE is taking off in data centers today -- just not as a server interconnect.
Whether 40 or 50GbE will be the next rate for server interconnects and how this ties into the 400GbE speed currently in development are good questions that need to be debated. As I said, never underestimate the need for industry consensus building. To that end, the Ethernet Alliance will be hosting its next Technology Exploration Forum -- TEF 2014: The Rate Debate -- on Thursday, Oct. 16, in Santa Clara, Calif. All these topics will be brought to the table there.
— John D'Ambrosia is chairman of the Ethernet Alliance, the former IEEE P802.3ba 40Gigabit and 100 Gigabit Ethernet Task Force, and the current IEEE P802.3bs 400GbE Task Force.