The Devil We Know
It's worth remembering that despite all of its flaws, all of its architectural kluginess and insecurity and the costs associated with patching it, the Internet still gets the job done. Any effort to implement a better version faces enormous practical problems: all Internet service providers would have to agree to change all their routers and software, and someone would have to foot the bill, which will likely come to many billions of dollars. But NSF isn't proposing to abandon the old network or to forcibly impose something new on the world. Rather, it essentially wants to build a better mousetrap, show that it's better, and allow a changeover to take place in response to user demand.

To that end, the NSF effort envisions the construction of a sprawling infrastructure that could cost approximately $300 million. It would include research labs across the United States and perhaps link with research efforts abroad, where new architectures can be given a full workout. With a high-speed optical backbone and smart routers, this test bed would be far more elaborate and representative than the smaller, more limited test beds in use today. The idea is that new architectures would be battle tested with real-world Internet traffic. "You hope that provides enough value added that people are slowly and selectively willing to switch, and maybe it gets enough traction that people will switch over," Parulkar says. But he acknowledges, "Ten years from now, how things play out is anyone's guess. It could be a parallel infrastructure that people could use for selective applications."

[Click here to view graphic representations of David D. Clark's four goals for a new Internet architecture.]

Still, skeptics claim that a smarter network could be even more complicated and thus failure-prone than the original bare-bones Internet. Conventional wisdom holds that the network should remain dumb, but that the smart devices at its ends should become smarter. "I'm not happy with the current state of affairs. I'm not happy with spam; I'm not happy with the amount of vulnerability to various forms of attack," says Vinton Cerf, one of the inventors of the Internet's basic protocols, who recently joined Google with a job title created just for him: chief Internet evangelist. "I do want to distinguish that the primary vectors causing a lot of trouble are penetrating holes in operating systems. It's more like the operating systems don't protect themselves very well. An argument could be made, 'Why does the network have to do that?'"

According to Cerf, the more you ask the network to examine data -- to authenticate a person's identity, say, or search for viruses -- the less efficiently it will move the data around. "It's really hard to have a network-level thing do this stuff, which means you have to assemble the packets into something bigger and thus violate all the protocols," Cerf says. "That takes a heck of a lot of resources." Still, Cerf sees value in the new NSF initiative. "If Dave Clark...sees some notions and ideas that would be dramatically better than what we have, I think that's important and healthy," Cerf says. "I sort of wonder about something, though. The collapse of the Net, or a major security disaster, has been predicted for a decade now." And of course no such disaster has occurred -- at least not by the time this issue of Technology Review went to press.