Ethernet and Data Gathering for HPC
In my 15 or so years leading the charge for Ethernet into higher speeds “high performance computing” and “research and development” have always been two areas that the industry could count on where higher speeds would be needed for its networking applications. For example, during the incarnation of the IEEE 802.3 Higher Speed Ethernet Study Group that looked beyond 10GbE, and ultimately defined the 40 Gigabit and 100 Gigabit Ethernet project, Figure 1 tracked the average processing against the prior introductions of new rates of Ethernet. For “research and development” a number of key areas were identified as needing terabit capacities from an end-to-end perspective, such as high energy physics, climate, nanoscience, fusion energy, astrophysics, and genomics data and computation.
Moving forward to 2012 and the IEEE 802.3 Ethernet Bandwidth Assessment Ad hoc, presentations for HPC and research and development were brought forward for bandwidth consideration. So, it is easy to see a synergy between Ethernet and the HPC and research and development application spaces in terms of higher speeds, but is there more to that relationship?
Before I answer that question, I would like to diverge for a moment and reflect on my experience in the Ethernet community. I have been fortunate to develop relationships with many senior individuals in the IEEE 802.3 Ethernet Working Group, as well as within the Ethernet industry, including spending a lot of time with Bob Metcalfe during preparations for the multiple 40th Anniversary celebrations we attended last year. What I have taken from all of these individuals is that one should never try to second guess the market, because people will figure out how to use Ethernet in ways probably never considered.
And it should be pointed out that these areas can grow into very high-volume applications. Need a couple of examples? The IEEE 802.3 has multiple efforts to provide 100 Mb/s to 1Gb/s Ethernet bandwidth solutions inside of cars, plus power. This application area is forecasted to ship in the 100s of millions of ports by the end of the decade. How about Power-Over-Ethernet (PoE)? PoE is another huge area that has grown to such a point that the Ethernet Alliance is now contemplating a logo certification program.
These two application spaces highlight a very important point, which is the role that Ethernet can play in gathering data for HPC and research and development. Why? Well, first, Ethernet technologies are becoming so low cost that they are moving into entirely new application areas. Next, in comparison to 100GbE and 400GbE, DC to 1GbE is fairly low bandwidth. Low cost — low bandwidth — lots of volume! This is what makes Ethernet a viable candidate for support of wired applications for the Internet of Things or “IoT” (as it is lovingly known in the marketing world). And let’s be clear — while I got excited before when talking about 100s of millions of ports, when we talk about IoT, which Gartner is citing will be 212 billion ports by the end of 2020, I get really excited.
Sensors will present an opportunity for information gathering on a multitude of problem areas that have never really been addressable in the past. Low bandwidth — low cost — tons of applications! And, just as my mentors have taught me, don’t try to second guess the market. I am not going to second guess the work that is being considered by individuals today to use Ethernet in ways that I never before considered.
 “Higher Speed Study Group Call-For-Interest,” July 2006,http://www.ieee802.org/3/cfi/0706_1/CFI_01_0706.pdf
 “IEEE 802.3 Ethernet Bandwidth Assessment,” July 2012, http://www.ieee802.org/3/ad_hoc/bwa/BWA_Report.pdf
John D’Ambrosia is chairman of the Ethernet Alliance and chief Ethernet evangelist, CTO office at Dell. He may be reached at editor@ScientificComputnig.com.