The Road Ahead
With students surfing the Web, downloading music files, and working on problem sets on remote servers, the network is running with real traffic. Now Morris and the four graduate students working with him full time on the project can test different routing strategies that better adapt to the hostile wireless environment.Their idea to cope with the unpredictable environmental disruptions is to figure out not just whether two nodes can hear each other, but also measure how well they can communicate. Instead of finding the shortest path between two nodes, their protocols try to find the best path-the one in which data packets won’t get stuck or corrupted along the way. This requires a constant monitoring of the links. Roughly once per second, each node sends out a small “hello” broadcast packet. All the other nodes record whether they receive this probe, keeping a history of the last 10 probes. So if, say, node A has sent out 10 probes and node B received 8 and node C received 4, then the routing software knows that the path A-B is better than path A-C. Also, every 15 seconds, every node sends a broadcast message that lists the nodes it knows how to reach-and the link quality for each associated path. That way, all nodes have a complete, continually updated, routing map of the entire network-and know the optimal routes for reaching one another.
In building Roofnet, the MIT researchers found many things they didn’t expect. For example, the range of the 802.11b cards and antennas vary considerably. “We’re now skeptical about what manufacturers say,” says John Bicket, one of the grad students working on the project. “We found nodes that couldn’t talk across the street, but others could talk half a kilometer apart.” The cause might be local environmental conditions or even multiple reflections of the same signal that cancel themselves out. Another surprising phenomenon is the lack of symmetry in the link transmission quality: it is not uncommon for node A to be able to send data to node B easily, while node B can’t reciprocate. Such anomalies complicate the development of routing schemes.
By debugging and fine-tuning their routing schemes, the MIT researchers hope they will be able to use them in even more complicated systems. One such situation would be when nodes are not static in rooftops, but moving at different speeds in all directions-a scenario not far in the future, as more and more people carry personal digital assistants and cars are beginning to be equipped with computers. “It’s a matter of tuning the protocol so that it can handle mobility,” says Sanjit Biswas, another student involved in the project.
Ultimately, Morris says the group plans to release the Roofnet routing software as a freely dowloadable open source program. That means that anyone with a computer and a Wi-Fi card would be able to install the routing software and become a node in the network. Other people in other areas could also download the software and create their own rooftop community networks.
Of course, many problems still need to be addressed. First, MIT can’t provide Internet access to non-MIT affiliates; the network would therefore eventually have to find other gateways to the fixed Internet. But that raises another complicated issue: most Internet service providers don’t want their users sharing their bandwidth. Also, the community networking technology needs to guarantee a certain level of security and privacy. With users literally sending their data through the air, via other people’s nodes, some sort of encryption will probably be necessary to avoid eavesdropping. It is also necessary to guarantee a fair, balanced used of the system, to avoid that a single user sucks all the bandwidth and clogs the network. Finally, the system needs to be robust enough to resist some more pragmatic problems-such as when snow forms in the antennas.
When the day comes, what will happen? Again, the MIT group wants to learn by doing. “We’ll see,” says Morris.