We consider the problem of controlling mobile data ferries for message delivery among disconnected, scattered domains in a highly partitioned network. Existing work on data ferry control mostly focuses on predetermined ferry routes, assuming full observations at the ferry and no explicit Quality of Service (QoS) constraints on the resulting communications. In this paper, we aim at designing a QoS-enabled ferry control solution, which handles both partial observations and bounded message delays. To this end, we extend our previous work on data ferry control with partial observations into a comprehensive hierarchical framework called Switch-and-Navigate (SAN), which consists of a global switch policy for determining the best domain to visit and a local navigation policy per domain for searching for nodes within individual domains. Under the assumption of Markovian node mobility, both the global and the local control problems are formulated as Partially Observable Markov Decision Processes (POMDPs) to maximize the discounted effective throughput over all domains. Due to the fact that the optimal solution to POMDP is PSPACE-hard, we develop heuristic policies and further approximations for efficient computation. Simulation results show that the proposed policies can significantly improve the performance over predetermined alternatives.
展开▼
