The Internet sky really is falling
- 07 May, 2009 08:30
Many folks are familiar with the modeling we've done over the past few years highlighting the fact that Internet demand is outstripping capacity, specifically access capacity. The findings were, to put it mildly, controversial: We've been called everything from carrier shills to nut-jobs. (No, the research wasn't sponsored. And we never claimed your fillings were receiving extraterrestrial radio signals).
The bottom line? We were right. YouTube recently announced it's discontinuing video delivery to certain geographies due to -- ahem -- lack of access capacity. And providers from telcos to cable companies are implementing "usage caps" to keep users from, er, consuming "too much" bandwidth. Seems the only thing we got wrong was the timing --we anticipated the crunch hitting in the 2011/2012 timeframe, but we're seeing it happening already.
Time for the really bad news. Access capacity shortage isn't the only -- or even the worst -- problem facing the 'Net. IP itself is nearing end-of-life, with no ready alternative. Pretty much everyone's aware that we're running out of IPv4 addresses at an alarming rate, and despite more than a decade of massive promotion, IPv6 deployments are a tiny fraction of what they would have to be to meet the gap. A few people are also aware that due in part to increased multihoming, routing table sizes are increasing dramatically, to the point where they'll exceed Moore's Law's ability to keep up. (IPv6 actually makes this problem worse, although how much so is a matter for debate).
As described in a paper presented at the January NANOG: For service providers, the Internet is about to become a lot more expensive to deploy and operate; for users, the Internet is about to become a lot less reliable and a lot more expensive (and balkanized).
It gets worse: There's no clear fix. Next-generation Internet projects have come and gone over the years, with little real success. Several projects are underway, but they're nowhere near complete -- and there's no consensus that any of them will actually work. The approach with the greatest momentum at the moment is Location/ID Separation Protocol (LISP), developed by some of the brightest people in the 'Net, and supported by Cisco.
Another is referred to as PNA, or Patterns in Network Architecture, after the eponymous book by its inventor, John Day. PNA is promoted by the Pouzin Society, named after Louis Pouzin, French inventor of the datagram, which held its first meeting recently at the FutureNet conference in Boston . There's also the Trilogy project, a European academic collaboration.
But none of these projects are far enough along to address the looming crisis. There's a prototype implementation of LISP, but nothing in production. PNA defines an architecture, but lacks an implementation. And Trilogy is in the relatively early stages of setting up collaborative working groups -- an actual architecture, much less an implementation, is a ways off.
I'm fairly confident the current challenges will be met, because there are enough bright minds concentrated on the problem, and at least one potential architecture exists. But buckle your seatbelts because there's likely to be turbulence ahead.