Untangling a Web: The Internet gets a new look

The Internet may be everywhere and nowhere, but that’s not stopping information engineers from mapping it. An atlas that accurately shows the physical path of information from one computer to another could protect the Internet from massive failures after, say, an earthquake or a terrorist attack. The latest finding by Internet cartographers offers some good news: Routing computers—the hardware that directs bits of information around the world—are linked in a way that makes the Internet less susceptible to a centralized attack than past studies had indicated.

NET WORKS. The Internet doesn’t have centralized master routers, according to a new analysis. Its distributed structure, as depicted here, protects it against a widely debilitating failure. PNAS

In an upcoming issue of the Proceedings of the National Academy of Sciences, John Doyle of the California Institute of Technology in Pasadena and his colleagues offer a new mathematical model of the Internet. The team used a class of models known as HOT, for highly optimized tolerance.

This approach differs from the so-called scalefree models that many Internet modelers have favored (SN: 1/29/05, p. 68: Sizing Up Complex Webs: Close or far, many networks look the same). Extrapolating from small local networks, the scalefree model indicates that a few well-connected master routers direct Internet traffic to numerous, less essential routers in the network’s periphery. The model predicts that over time, the busy central routers acquire ever more links in a rich-get-richer effect. From a security perspective, a targeted attack on a central router could halt virtually all data flow in such an Internet.

Doyle and his team present a HOT model in which the Internet has no vulnerable centralized hubs and any highly connected routers lie at the periphery. If one of these well-connected, outlying routers were taken out, Internet traffic would simply divert to another well-connected router.

That approach incorporates current technological and economic constraints on the Internet. For instance, routers are limited in the amount of information that they can direct at any given instant. Additionally, Internet-service providers that own major routers can allocate only limited resources to their hardware. By optimizing these factors in their model, Doyle and his coworkers aim to present a picture that matches today’s Internet.

The team has tested its model with the known map of Internet2—an academic subnetwork within the larger Internet. The researchers report that their proposed model corresponds well to the structure of Internet2, which Doyle says is a “good representation” of the structure of the entire Internet.

Identifying the strengths and weaknesses of different Internet architectures is important, says Fan Chung of the University of California, San Diego. She notes that HOT models can also be used to analyze other complex networks, such as the immune system and protein interactions.

Although the researchers’ model reveals that the Internet ought to efficiently adapt even if the busiest routers were removed, Doyle notes that the system is still vulnerable to hijacking via malicious software that leaves the hardware intact. Even in that case, however, a good map could help Internet custodians track harmful software and ultimately protect the Internet against it.

More Stories from Science News on Computing

From the Nature Index

Paid Content