부산대학교 도서관

In recent years the sensor network databases TinyDB and StonesDB have emerged. While both provide a useful abstraction layer for querying data, live data in the case of TinyDB and historic data in the case of StonesDB, neither of these approaches provide transaction processing capabilities. Transaction processing capabilities are needed to guarantee the consistency, for instance, in the case of data updates at runtime or the sophisticated requirements of emerging wireless sensor and actor networks. In this paper we analyze traditional concurrency control approaches and compare them with regard to their usability in wireless sensor networks. Therefore, we implemented the traditional Strict Two Phase Locking (S2PL), Timestamp Ordering (TO) and Forward Oriented Optimistic Concurrency Control (FOCC) by validation. We show in experiments with the network simulator Shawn, that locking exhibits better commit rates and lower costs under a variety of conditions compared to timestamp ordering and validation. We also implemented locking for the sensor node platform Pacemate to show the feasibility of our approach. We believe that efficient concurrency control can broaden the application spectrum of sensor network databases and is also vital for the emerging wireless sensor and actor networks.