Fair and efficient scheduling in data ferrying networks

Data-ferrying disconnection-tolerant networks allow remote rural areas to access the Internet at very low cost, making them viable alternatives to more expensive access technologies such as DSL, CDMA, and dial-up. In such a network, an Internet-based proxy gathers data from the Internet and sends it to a set of edge nodes called "gateways", from which data ferries, such as buses and cars, opportunistically pick up the data using short-range WiFi as they drive past, and deliver it wirelessly to kiosks in remote villages. In this context, we pose the following question: assuming knowledge of ferry schedules, when and to which gateway should the proxy send each data bundle so that the overall delay is minimized and the bandwidth is shared fairly among competing kiosks? We show that a well-known schedule-aware routing scheme proposed in the literature, i.e., EDLQ [11] is far from optimal. Moreover, EDLQ does not provide means to enforce bandwidth allocations. To remedy these problems, we employ a token bucket mechanism to decouple fairness and delay minimization concerns. We also describe a utility-maximizing scheduler based on the classical minimum-cost network flow problem, that finds optimal schedules. Through simulations, we show that our scheme performs at least as well as EDLQ in scenarios that favour EDLQ, yet achieves up to 40% reduction in delay in those that do not.