Stream Programming for Distributed Systems
Andrew Santosa, University of Sydney
Date and Time
Thursday, October 18th, 2012 at 10:30
Polacksbacken, room 1112
Wireless sensor networks (WSN) and mobile clouds are composed of sensor nodes that have limited energy resources. For wireless sensor networks, query processing is the state-of-the-art for data gathering and processing applications to avoid low-level programming. The stream programming model has been widely used to represent queries as an information flow from the sensor nodes to the base station. The model describes queries as stream graphs consisting of operators that process data and channels that connect operators. Operators are deployed in the network to reduce the communication overhead and hence energy. The modification of WSN queries at runtime is of key importance due to changes in the environment and the network energy levels, resulting in the migration of operators between the network nodes.
In this work, we introduce the migrating operator placement problem (MOPP) that places operators of stream graphs on sensor nodes, such that energy costs are minimized. The placement takes changes of queries and migration of operators into account. The general MOPP is NP hard, and, therefore, we develop a dynamic program for a compositional subset of the stream graphs with polynomially-bounded running time. To improve the performance of our algorithm, we introduce a heuristic that reduces the search space to the proximity of the base station. We conduct various experiments using a simulator for wireless sensor networks with different sizes.