Adaptive and Decentralized Operator Placement for In-Network Query Processing

Abstract

In-network query processing is critical for reducing network traffic when accessing and manipulating sensor data. It requires placing a tree of query operators such as filters and aggregations but also correlations onto sensor nodes in order to minimize the amount of data transmitted in the network. In this paper, we show that this problem is a variant of the task assignment problem for which polynomial algorithms have been developed. These algorithms are however centralized and cannot be used in a sensor network. We describe an adaptive and decentralized algorithm that progressively refines the placement of operators by walking through neighbor nodes. Simulation results illustrate the potential benefits of our approach. They also show that our placement strategy can achieve near optimal placement onto various graph topologies despite the risks of local minima.