^{Democracy & Technology Blog} A Breakdown of the Innovation Culture?

In preparation for the “Exaflood” paper, I read the November 2007 paper by Nemertes Research — “The Internet Singularity Delayed: Why Limits in Internet Capacity Will Stifle Innovation on the Web.” It is an exemplary supply-side work (low utilization rates signify inadequate bandwidth rather than lack of demand). Failure to invest in infrastructure will produce not a breakdown of the Internet but a breakdown of the innovation culture of the net that brought us YouTube et al.
I recommend the paper to all as a guide to the prospects of our network processor and hollow router paradigms. It contains a number of obvious errors (dates reversed on charts (p.22), confusions between zettabits per second and petabits), and a “What me worry?” approach to huge conflicts between Nemertes and Odlyzko estimates of global capacity in 2000 (Odlyzko 85 pettabytes per month; Nemertes 61 exabytes!). Today global access capacity is around a zettabyte (10 to the 21) per month (2 plus petabits per second), with the U.S. commanding only one seventh of it (300 Tbps) or 14% while our GDP was close to 20% and our market cap 40% (adjusting for dollar doldrums).
Meanwhile, U.S. investment in infrastructure (capex only) was roughly $5B out of a global total of $20B and U.S. investment in access equipment (again capex only) was under $1B or about a fifth of global access investment in capex ($5B plus). But the U.S. out-invests the rest of the world in edge router/switch connectivity for the metro and high-end enterprise. Enterprise IP traffic is estimated to be about 1.5 times Internet IP traffic, but convergence continues.
On page 29, the report contains a breakdown of core and edge router/switch unit growth that is relevant to our network processor paradigm. On the order of 10 to 15 thousand core routers are sold annually, compared to between 30 and 60 thousand edge and metro routers and literally billions of access nodes of all kinds. The NPA is a key product for volume production of NPUs for scale and learning curves.
The other insight is that IPV6 meets the need for addresses but does not respond to the expansion of router tables that will slow the net in coming years if it is not remediated. The conclusion is a large need for CAMs, Knowledge Processors, and other memory and lookup table accelerators.
Nemertes declares that the U.S. confronts a coming bandwidth crunch in 2010, when access constraints will begin seriously to limit investment in Internet service innovation. The argument is that Moore’s law increases in capacity will yield rising utilization rate and that traffic is supremely sensitive to utilization rates. In other words, as bandwidth increases we use it more and innovate more. I believe this. Others don’t.
“If we build it, they will come,” is the underlying assumption of Nemertes and me. People laugh but it is true over any run longer than a year or so.