400Gb Ethernet
April 3, 2013, Ethernet Technology Summit, Santa Clara, CA—John D’Ambrosia from Dell and the Ethernet Alliance described the efforts to develop 400Gbps Ethernet. These efforts are the start of something that will lead to improvements for everyone.
The industry is getting involved in early standards developments and research. The ecosystem of Ethernet will change to accommodate all the ranges of speeds and interconnections. The increasing number of users and devices demands more speed and higher data rates will lead to even more content.
We have already seen that 40G and 100G systems are enabling new services. All the new data rates will eventually move into the data center. By assessing bandwidth sources and growth rates, we see that average rates are doubling every 18 months, a 58 percent CAGR. We will need Tb capacity between ’15 and ’20, and already we know that some traffic is running at over 2.5Tb/s.
The standards bodies are looking at common issues so the IEEE and ITU are starting meetings. The challenge will be in the economics areas and not in the technical specifications. The cost of the TbE will dictate the solutions. The cost of moving a bit has to go down to allow a reasonable tradeoff between speed and reach at the PHY level.
The modulation scheme will matter, as will other technical issues. The increasing complexity dictates that we will need parallel development flows for the different PHYs. We need to define the architecture so the data center interconnect and reach issues can be addressed.
The first target will to reduce the number of connectors, so a multi-mode operation will have to exceed the current 100 m reach. We do not have any standards or technologies for the 150 to 1km reach, and the 500m reach will be very important, as this is a corner to corner connection in the data center. In fact, we increased the reach for the 40 G specification for that reason. Single mode fiber is capable of 300m-2 km reach, so it is the likely candidate for 400GbE.
400Gbe is going to be needed everywhere, so the technologies need to catch up to the requirements. The architecture will have to address many layers and all will have to work together. Flexibility and scale will be required, but we also have to balance cost, power, and density. The options for 400GbE are to reduce the number of signals to reduce cost and power, and some type of electrical to optical conversion. The most promising technologies are in the III-V materials or in silicon photonics integrated into the circuitry. At least one will have to work over time.
The interface width is likely to be a x16 module, for 16 x 25GbE. The alternative is to go to 8 lanes of 50Gbps. We have 8-10 years for adoption after the specifications are settled, so there are many architectural variants possible and many of them will be in use. Multi-rate systems will trade speed with density, so we will have to change the building blocks and allow different languages per segment. The first specification will be ready about 2017 and may include embedded optics.
