A Location Based Aggregation Algorithm for Selective Aggregate Queries in Sensor Networks

In-network data aggregation algorithms are based on the premise that the energy requirements for data collection in sensor networks (SNs) can be significantly reduced by aggregating and collecting individual sensor readings over an efficient data collection path. In this work, we focus on selective aggregate queries, i.e., queries that aggregate data only from a subset of all network nodes. The task of optimal data collection in such queries is an instance of the NP-hard minimal Steiner tree problem. We present an aggregation algorithm, called Pocket Driven Trajectories (PDT) that optimizes the data collection path by approximating the global Steiner tree with a minimal overhead using purely local spatial knowledge. We show that selective aggregate queries can lead to various node participation scenarios characterized by spatial factors such as the distribution of participating nodes over the network, i.e., clustered vs. dispersed, location of clusters, inter-cluster dispersion, location of the sink with respect to the participating nodes, and location and size of communication holes. Our experiments compare the performance of well-known in-network aggregation algorithms against PDT in partial node participation scenarios. A globally approximated minimal Steiner tree serves as a benchmark for all of the aggregation algorithms. We show that PDT (a) leads to considerable gains in selective aggregate queries and (b) provides a close approximation of the minimal Steiner tree.