It's hard to believe that it's been 10 years since David Isenberg published his seminal paper, "The Rise of the Stupid Network."
For those who don't know, here's the story: Isenberg, then employed as a researcher at AT&T's Bell Labs, decided to challenge some key networking premises. Specifically, he attacked the notion that networks should be designed around highly intelligent centralised control.
Instead, he proposed that networks should be comprised of dumb transport in the middle, with intelligent user-controlled endpoints. His employer -- which had invested millions of dollars developing highly intelligent centralised control systems -- disagreed with that perspective, and they subsequently parted ways.
It seems obvious that Isenberg was absolutely right, given the stratospheric success of the Internet (the ultimate in "intelligent user, stupid networks"). But if intelligent networks are obsolete, what's behind the current vogue for "application-aware networking"?
The most recent example is AT&T's just-announced "application acceleration" service, but whether it's Cisco's service-oriented network architecture (SONA), Juniper's application acceleration, or the deep-packet inspection by these vendors and others (Nortel), the current trend is to equip network routers and switches with the ability to make on-the-fly decisions about how to handle traffic.
In other words: Isenberg's dumb network is getting smarter.
Now, you could argue that adding intelligence to routers and switches isn't exactly the same thing as creating "highly-intelligent centralised control." But that's hair-splitting. At the end of the day, application-aware networking is all about building intelligence back into the network.
How bad an idea is that? Keep in mind that Isenberg wasn't so much promoting a "network stupidity doctrine" as searching for a provocative way to highlight the emergence and viability of highly distributed IP-based networks. On that point, events have shown him to be 100 percent correct.
But network architecture always embodies a fundamental tension between capability and control. "Smarter" networks limit what you can do in the hopes of delivering guaranteed performance. "Dumber" networks offer a sky's-the-limit suite of choices, but limit performance guarantees.
In other words, if a network were a political/economic system (which in some respect it is), the tension is between the stability of a dictatorship and the anarchy of a free-market democracy. The lesson of the Internet (and of free-market democracy, at least so far) is that more freedom is generally preferable, even at the cost of limited performance guarantees. But "generally preferable" doesn't mean "true at all costs" -- nobody's yet been able to build a fully sustainable anarchy.
In network terms, the network should be dumb enough to permit freedom, but smart enough to stay functional under stress. How dumb is dumb enough? It reminds me of the adage of how to determine the carrying capacity of a bridge: Keep driving trucks over it until it collapses. In other words, add just enough intelligence to keep the 'net functional, but not so much that it breaks.