A lot of excitement is brewing in the electronics industry about the so-called Internet of Things and the huge market opportunity it represents: 50 billion devices connected by 2020, according to some estimates. But add that traffic to the dramatic rise in wireless networking and entertainment streaming, and you have a problem. The backhaul infrastructure is going to clog unless something is done, quickly.
The potential for a clogged network is no surprise to some. Last year's "IEEE Ethernet Bandwidth Assessment Report" spelled things out clearly. The report expects a 100-fold increase in network traffic by 2020, requiring networks to support up to 10 Tbit/s. The report concludes that the network will need 400G link speeds (or their equivalent in parallelism) as early as 2015.
With the need thus clearly spelled out, the IEEE has begun the first steps toward creating a 400G Ethernet standard. A 400G study group was formed in April this year and has already had its first meeting. The next is set to occur in July. The meeting materials from the first study group gathering are now online.
Clearly there is a great deal of work to do in defining a new standard, not the least of which is determining the underlying architecture. Will the link use ten 40G lanes in parallel? How about four 100G lanes? What makes the most sense in terms of implementation cost and complexity?
One of the arguments made in the presentation, "Scalable 400GbE Architecture" by Ali Ghiasi of Broadcom, is that adopting the approach of sharing the 400G PMD (physical medium dependent) sub layer with 100G Ethernet links could help speed deployment of the technology. The approach is similar to one taken in leveraging 10G Ethernet to implement 40G.
Ghiasi estimated that there has been a four-year lag in the ramp-up of 40G and 100G port shipments. A similar lag in 100G to 400G shipments would push the volume deployment of equipment out beyond 2020 -- much later than the Bandwidth Assessment report estimates it will be needed. Following the 4x100G approach could speed the process, Ghiasi stated.
There are alternative suggestions on the table. One possibility is to use a 4x25G structure for installing 100G ports, and then use a 16x25G structure, technically feasible today, for 400G. Or perhaps an 8x50G. The 10x40G approach seemed to be favored by the attendees.
Whatever approach finally gets settled on, it seems essential that the choice be made with a strong bias toward what is within the grasp of today's technology (no breakthroughs required). The need to increase capacity rapidly is beginning to become pressing, and the industry needs a clear standard on which to build. 400G Ethernet needs to be on the fast track.
For the moment, anyway, it seems to be. In the pre-meeting materials for the July gathering is a strawman Project Authorization Request (PAR) for development of a 400G Ethernet standard: IEEE 802.3bs. Completion and approval of the PAR are the next steps needed in the run towards 400G.
What are your thoughts about 400G Ethernet?