Most of the routing protocols now being proposed by mesh network researchers borrow the shortest-path strategy used in the fixed Internet. These protocols try to find the route with the fewest number of intermediate nodes between sender and destination. For the wired Internet-with its nearly static topology and reliable links-this scheme has been working pretty well: our e-mails hop from router to router and reach the other side of the world in a few seconds.But it turns out that this shortest-path strategy might not be adequate for sending packets through the air. In a wireless network, according to the MIT group, distance matters: the longer the signal has to travel, the more it will degrade. Moreover, the link quality between nodes varies unpredictably due to such transient phenomena as trucks driving by, moisture in the air, or a pigeon sitting on the antenna. The result is a considerable amount of packet loss, transmission errors, and connections that simply appear and disappear throughout the day. A routing protocol that minimizes the number of hops ends up choosing longer distances for each hop-and therefore sending data over low-quality wireless links.
The MIT group realized that new routing strategies were necessary when they deployed an initial version of Roofnet last spring. They tried to implement some of the proposed routing protocols discussed by the Internet Engineering Task Force, the organization governing the Internet’s technical standards. But while these protocols work well in theory-and are generally tested in computer simulations or small-scale, laboratory networks-they don’t take into account many unpredictable factors involved in radio communication. The protocols usually assume, for example, that when one node can detect one nearby, it can communicate well with its neighbor. But that turns out not always to be true. The MIT researchers and other groups have found that many times two nodes can “hear” each other by exchanging small probe packets, but when they try to send real data, communication collapses due to inadequate bandwidth. Morris and his group decided that the best way to develop robust wireless routing protocols was to test them with a real network, real users, and real traffic.
Other mesh network researchers say the MIT work represents an important advance for debugging these routing schemes. “Their work is grounded in real system building,” says Victor Bahl, a senior researcher who leads the networking group at Microsoft Research in Redmond, WA. “The insight that you get out of building things is a lot more than you’ll ever get if you just simulate things.” Demonstrating that such networking is viable in a real, large-scale implementation, he says, is a crucial step toward attracting more industry attention to the technology’s potential.
Deploying such a network became possible because Wi-Fi technology has gotten so cheap. A few years ago, Morris says, the price of the wireless cards would have made the project prohibitively expensive. Each Roofnet node uses an 802.11b wireless networking card installed on a cheap PC running Linux and the routing software. A coaxial cable connects the wireless card to an omnidirectional antenna. The user then connects the PC to the Roofnet node. The total cost of the equipment for each node is $685.
To deploy the network quickly, the MIT group distributes free self-installation kits to students who want to participate in the project. For these students, getting the Roofnet node running is part of the fun. “Our antenna was put up by a friend of mine who does rock climbing,” says graduate student Roshan Baliga, who lives in a two-story building with no easy roof access. “He scaled the side of the apartment to get to the roof, installed the antenna, and then rappelled down.”
MIT students are happy to participate in the project, especially because they can save some money. “We compared a broadband cable connection to Roofnet and couldn’t tell the difference, so we cancelled the cable,” says MIT senior Walt Lin, who installed the antenna on his sloped roof.