Harvesting Time-Series Data from Service-Based Systems Hosted in MANETs.

We are concerned with reliably harvesting data collected from service-based systems hosted on a mobile ad hoc network (MANET). More specifically, we are concerned with time-bounded and time-sensitive time-series monitoring data describing the state of the network and system. The data are harvested in order to perform an analysis, usually one that requires a global view of the data taken from distributed sites. For example, network- and application-state data are typically analysed in order to make operational and maintenance decisions. MANETs are a challenging environment in which to harvest monitoring data, due to the inherently unstable and unpredictable connectivity between nodes, and the overhead of transferring data in a wireless medium. These limitations must be overcome to support time-series analysis of perishable and time-critical data. We present an epidemic, delay tolerant, and intelligent method to efficiently and effectively transfer time-series data between the mobile nodes of MANETs. The method establishes a network-wide synchronization overlay to transfer increments of the data over intermediate nodes in periodic cycles. The data are then accessible from local stores at the nodes. We implemented the method in Java~EE and present evaluation on a run-time dependence discovery method for Web Service applications hosted on MANETs, and comparison to other four methods demonstrating that our method performs significantly better in both data availability and network overhead.