In 1997, then-Bell Labs researcher David Isenberg wrote a groundbreaking story criticising AT&T's top-down approach to networks, which emphasised the importance of intelligence within the network. In his piece, Isenberg made a cogent argument favouring network "stupidity" - networks that focused simply on transporting bits, leaving the intelligence to end-user devices.
The story - which led to Isenberg's much-publicised departure from his then employer - neatly expressed the paradigm shift that had been under way for the previous couple of decades, and which was a classic case of the emergence of a disruptive technology: the IP revolution. (If you're not familiar with the concept of a disruptive technology, check out Clayton Christensen's outstanding book The Innovator's Dilemma.)
Before IP, networks were seen as top-down systems that required centralised control and end-to-end intelligence. The paradigm-shifting insight of the developers of IP was that if the network elements simply focused on switching packets, the needed intelligence could reside in the endpoints.
As history has shown, Isenberg's insight was correct. Not long after the piece appeared, his erstwhile employer committed itself to migrating toward an all-IP-based infrastructure (it now operates one of the largest and most sophisticated Internet backbones). Bottom line: IP and Isenberg won, the "intelligent network" lost.
Or did it? One of the great things about paradigm shifts is that they work for a while, but then they stop working. Lately I've been seeing a lot of signs that the packet paradigm has become less of a groundbreaking new vision than increasingly restrictive dogma - and whenever insight and innovation becomes unthinking dogma, it's time for a new paradigm.
A great example is the fervent pushback by packet purists against the use of MPLS. What's the problem with MPLS? In a nutshell, it's bad because it returns some control to carriers. But the great thing about MPLS, from a carrier's perspective, is traffic engineering. It allows carriers to determine paths across networks, thus imposing routing onto networks. To packet purists, this is unacceptable: The basic premise behind routing, after all, is that routers determine routes.
The problem is that routers are inherently non-deterministic, which means that without MPLS (or something very much like it), service providers have no way to offer consistent end-to-end QoS. And however much the packet purists may object, end-to-end QoS is something that end users happen to want.
MPLS is just one example. Stand by for other examples of cases in which the IP dogma has become, well, just a shade too dogmatic - and what that ultimately means for the next network paradigm.
Johnson is president and chief research officer at Nemertes Research, an independent technology research firm.