Article provided by EIU
Academics propose a new way to route data across the internet
From The Economist, October 3 2011
Imagine being in your seat on a flight from Houston to Manchester. After the plane takes off, the pilot announces that he has just been radioed to modify the flight path—through Beijing, say. The pilot doesn't know who told him to change course, but he follows his instructions anyway. On the layover in Beijing you might be left alone. Alternatively, you may be frisked and all your documents photocopied. Either way, it is all rather tedious.
Of course, airlines and passengers would never stand for such treatment. Yet this is precisely how data packets shuffle between networks that make up the internet. Fortunately, computer scientists at Carnegie-Mellon University (CMU) have come up with a clever way to eliminate the flaw. What is more, their solution could be introduced piecemeal in a manner that would not require a co-ordinated effort by everyone connected to the vast global network.
At present, there are over 200,000 independently operated networks, known as autonomous systems, which know how to shunt packets around the internet. Getting from a Comcast network in Houston to a BT network in Manchester, for instance, might require three or four intermediate hops. Autonomous systems constantly receive new routing instructions from the others—without much in the way of verification. That means that each autonomous system has to carry a constantly updated list of over 200,000 entries, where an individual entry contains the next stop required to move data from that autonomous system to one of the others.
Autonomous-system routers can typically choose from multiple paths to dispatch data packets to their destination. They may choose the cheapest, fastest or least congested. As a result, malicious, incompetent or accidental changes to the table of router entries can wreak havoc by, for example, letting mischief-makers hijack information. Anyone running a gateway that talks this routing language can publish incorrect paths, intentionally or otherwise. In April 2010, for instance, China Telecom told other routers to shift roughly 10% of the internet's pathways through its kit. The incident lasted about 18 minutes; the volume and nature of the re-routed data remain a mystery. At the time China denied this was a "hijack", but many observers were not satisfied with its explanations. Two similar events involving China have occurred since then. One in March affected just paths leading to Facebook.
The CMU researchers' proposal is known as SCION (which expands, rather ponderously, to Scalability, Control and Isolation on Next-generation Networks). The details may be abstruse, but the general idea is relatively straightforward. To start with, says Adrian Perrig of CMU, you need to create explicit domains of trust. Any independent network would be free to choose which other networks it wants to trust in a close embrace. Such high trust would not, however, be transitive. In other words, if a network belonged to two groups, its neighbours in one of them would not automatically trust its neighbours in the other simply by dint of having a mutual member. According to the proposal trust domains would comprise networks that operate under the same law and are bound by some form of binding agreement. That way, Dr Perrig notes, if a breach of trust occurred, there would be legal recourse.
Each time a member network sought to connect to another one in the trusted domain, it would still have to present a cryptographic signature as proof of identity. So far, that just seems like a method to secure groups of networks. Where SCION becomes more useful, and reduces complexity by an enormous degree, is when a router on a network wants to transfer data outside its closely trusted grouping.
Rather than remembering 200,000-odd entries, each one corresponding to each of the other networks, a network router in a trusted domain knows just a couple of facts. First, how to reach the very top of the routing hierarchy at the internet-transit service the network's owner subscribes to. That allows a network to rely on its transit provider to reach anywhere else in the world. Second, replacing those hundreds of thousands of entries are just a handful of pathways that tell a router how to pass packets along to a small number of neighbours which a domain has agreed to trust. It is as a pilot taking off from Houston would only know a handful of trusted airports at which he could land en route to Manchester. He would have no directory of all worldwide airports to consult, and the air routes to get there. On landing at the layover, he would plot the next step to Britain.
The trusty neighbours might be all the networks that agree to interconnect at a given network exchange point, or networks located far apart but which have close affinity, such as academic institutions, which have established direct links that bypass the public internet. The transit providers would then ensure that the connection is safe by using their own trusted routers. If an outsider tried to convince a trust domain to route information via an unknown path such requests would simply go unheeded. The new system would ensure that traffic meant to pass between trusted domains never went astray.
Such a structure can also be extended to make specific and fixed paths across the internet between networks, allowing virtual internets to be securely built without disturbing the existing networks. In the current set-up it is impossible to know precisely how packets will flow, Dr Perrig says. SCION would provide a clearer picture. And trusting only one's select neighbours prevents accidental network disruption. A typo in a (manually pecked) entry for a set of routes, for example, would not have the potential to take down large portions of the internet, route traffic to dead ends or overload equipment—all of which can happen today. "Even if you make a mistake in SCION, you won't mess up other peoples' traffic," assures Dr Perrig.
SCION requires much less computational power to operate, because a gateway using its approach need only look at a handful of entries, rather than analyse hundreds of thousands. Dr Perrig says this would reduce the cost of kit, and make SCION feasible for optically switched networks in which current route computation is just not possible. SCION can also be run in software without specialised chips.
The system is currently in the lab. Even if it got out, it is far from assured that it would be widely adopted. Many internet overhauls have been floated to little avail. That said, some require building a brand-new network infrastructure. SCION could be run over the existing internet. Independent networks could add SCION gear and fire up SCION links among trusted neighbours one route at a time.
The real question, though, is whether router manufacturers would embrace the change. SCION requires less powerful, and thus less expensive, hardware to operate. Makers of such hardware might not find that too appealing, even if their customers do.