An interview with Larry Roberts, one of pioneers of the Internet, was the highlight of this year’s Ethernet Technology Summit, a small, but very interesting event held in the heart of Silicon Valley.

Larry Roberts is best known as the leader of the team that created the ARPANET, a predecessor of the Internet, using packet switching techniques. The interview touched on his work on the first connections between computers in the US back in 1965, the initial challenge of promoting computer connectivity, and then keeping up with the demand once email’s popularity started to grow in 1970’s. Trying to get AT&T interested in offering email as a service in 1973 was, ironically, not very successful.

We often talk about how fast the technology is progressing, but Larry gave a few examples of how stubborn the industry can be. Some of the early definitions of Ethernet packet size, header format and addressing remain largely unchanged since 1970. In retrospect, many of these conventions are far from perfect, but they still work, and have drastically changed the world we live in. Is this a hint for the industry locked into so many debates today?

The interview was an incredibly modest account of some of the challenges met, mistakes made, and memories saved from the early days of the Internet, rather than how much all this work impacted the world we live in. It was a refreshingly non-marketing oriented speech, a tone shared by many other presentations at this unique, slightly old-fashioned, but very informative event.

John D’Ambrosia, Chairman of the Ethernet Alliance and Chief Ethernet Evangelist in the CTO Office at Dell - cheered up the attendees with his dynamic keynote presentation on the evolution of Ethernet and the challenges in coordinating development of IEEE standards, when the industry is being pulled apart by so many different opinions.

The root of this problem is diversity in demand for faster Ethernet connectivity across the industry. The leading Web2.0 companies are ready to deploy 100GbE and even 400GbE, as soon as a few vendors can supply it, without waiting for any standards or even industry consensus on the technology. On the other side of the spectrum are numerous enterprise campus network operators, deploying next generation Wi-Fi connectivity on campuses, who are just starting to look beyond 1GbE and realizing that 10GbE connections are too fast for the vast base of installed Cat5 cable. The result is new standards for 2.5GbE and 5GbE for campus networks, while datacenters are accelerating the introduction of 25GbE server connections as a base for wide adoption of 100GbE and 400GbE connectivity.

In-depth analysis of Ethernet evolution and an updated forecast for Ethernet optical transceivers will be published by LightCounting in June 2015, as part our updated 40G and 100G Data Center Interconnects report. The new title for this report may look like “25GbE - 400GbE and everything in between: Ethernet optical transceivers for Data Centers.”

Flexible Ethernet was one of the topics in the great plenary presentation given by
Brad Booth - Principal Architect for Azure Networking at Microsoft. Speaking about long distance connectivity between Microsoft datacenters around the world, Brad suggested that 250GbE or 350GbE may be just the right data rates for some of these very expensive 3,000 km connections. Paying for just enough bandwidth makes good economic sense.

It is interesting that at OFC 2015 Jeff Cox, senior director of Azure Networking at Microsoft, was talking about 100Tbps (yes, Tbps not Gbps) connectivity between data centers spread across a metro area in order to make them run as a single distributed datacenter. Apparently demand for bandwidth between datacenters separated by 3,000 km is more modest. Latency of these long distance connections is prohibitive for many applications: it takes 15 milliseconds for the light to travel over 3,000 km of fiber. In fact, the latency of 100 km connections across a large metro area - 0.5 milliseconds - may also be too high for many applications.

Latency of Ethernet switching fabrics was one of the key topics discussed at ETS 2015. Mellanox, one of the major sponsors of this event, is leading the industry in this category with latencies below 1 microsecond, equivalent to transmission over 200 meters of fiber. Will the speed of light ultimately limit the size of mega-datacenters or at least the size of colocation units also known as PODs?

A major trend discussed by many speakers at the event was the change in the industry supply chain driven by leading mega-datacenter operators.

News from Lightcounting...