Enhancing the Performance of Data Aggregation in Sensor Networks

Data aggregation is used for data-reduction in sensor networks and can dramatically reduce the amount of network communication, reduce their consumed energy and increase their throughput. But continuously computing data aggregates (such as MAX, MIN, AVG, SUM and COUNT) increases the likelihood of failures. These failures can be due to communication or node failures. Ignorance about failures can endanger the reliability of data aggregation computation. Existing approaches for detection of such failures, such as TAG and synopsis diffusion, have improved the reliability of the accuracy of computed aggregates, but suffer from weak performance on robustness of failures. These approaches have only used one member, i.e. parent node along a tree of nodes, for detecting failures. In contrast, in this paper we use the parent nodes as well for detecting failures. We propose an explicit approach that nearly eliminates the computational errors of aggregates that could arise due to communication or node faults. Evaluation results show that our proposed approach can improve the robustness of failures and the confidence on the correctness of computed aggregates