To bring data services through low-cost Internet access to remote rural areas, several projects proposed to use ferries, such as buses and cars,equipped with short-range WiFi to deliver messages between the Internet and remote villages. In such store-and-forward, ferry-based delay tolerant networks (DTNs), the messaging capacity on the ferries becomes a critical resource and thus needs to be used properly in order to reduce the overall transit delay. However, simply optimizing the overall transit delay may result in prolonged delays for certain individual villages. In this paper, we aim at ensuring a "fair" message ferrying service (in terms of transit delay) amongst the served villages while optimizing the overall performance of the whole system. To achieve our goal, we develop an application layer switch for message/ferry scheduling. Our scheduling is based on the notion of benefit that measures the advantages of ferry scheduling selections in a relative fashion. This approach allows us to properly model the delay minimization problem and fairness goal as mathematical constraints of a schedule optimization problem. Furthermore, we transform the two conflicting goals of this complex optimization problem into two interacting components, namely, performance optimization block and fairness assurance block, and treat them with bipartite matching and stochastic approximation