In the last years service oriented architectures have made their way from business applications to wireless sensor networks. For the successful usage of service oriented architectures, techniques like replication and migration of services are a prerequisite. We outline use cases for service migration and show that transaction processing capabilities are necessary for enabling stateful and consistent service migration in wireless sensor networks. Additionally, transaction processing capabilities are needed for the extension of sensor network databases like TinyDB and StonesDB and for wireless sensor and actor networks. In this thesis we compare the Two Phase Commit protocol and the Cross Layer Commit Protocol with regard to their usability in wireless sensor networks and infer our own atomic commit protocol Two Phase Commit with Caching. We show in simulations and experiments with the sensor node platform Pacemate, that our Two Phase Commit with Caching protocol is the most efficient protocol for use in wireless sensor networks with regard to energy consumption. Further we implement and evaluate the traditional database concurrency control protocols locking, validation and timestamp ordering for wireless sensor networks and show in simulations that locking outperforms validation and timestamp ordering in terms of efficiency. We also report our implementation of locking for the Pacemate sensor nodes. We describe a comprehensive service migration scenario, which enables service discovery and consistent migration of stateful services and we outline our implementation for the network simulator Shawn and Pacemates running Surfer OS. Finally, we describe an adaptive selection of commit protocols and routing protocols in dependency of the required consistency and the given network context to provide the most efficient transaction processing for a given sensor network deployment. This thesis supports the usage of sophisticated applications for wireless sensor networks demanding increased coordination capabilities while taking the severe resource constraints of wireless sensor networks into account.
|Doctorate / Phd
|Published - 21.06.2011