Performance Analysis of Data Aggregation Schemes for Wireless Sensor Networks

Abstract

Wireless sensor networks are suitable for applications in which sensors detect moving targets in the area of interest. In such applications, one of the key challenges is how to design efficient data aggregation protocols which reduce redundant packet transmissions in the sensor network. Towards this goal, centralized, tree-based, static-cluster, and dynamic-cluster aggregation schemes have been proposed. However, each scheme has its share of benefits and corresponding costs, and it is difficult to say with certainty whether a particular scheme is always better than others. In this paper, our goal is to compare all of the above mentioned aggregation schemes with comprehensive theoretical analysis, simulation and real experiments and attempt to give clear scenarios where a particular scheme may be more beneficial compared to others. Along the way, we also propose two cluster-based aggregation algorithms, which are simple enough to be implemented on resource-constrained sensor networks. As a first step towards this goal, we model the sensor network environment under certain simplifying assumptions and then derive closed form expressions for the total number of packet transmissions incurred by each aggregation scheme. Next, we complement the assumptions made for the analysis by performing extensive simulations under different environmental conditions, such as channel capacity (1Mbps, 250Kbbps) and MAC (B-MAC, IEEE 802.11) protocol. Finally, we test the aggregation schemes on a real sensor network testbed comprising of 31 Mica2 sensors. Various metrics, such as total number of packet transmissions, aggregation ratio, average energy consumption, network lifetime, average end-to-end delay, and packet delivery ratio, are used to evaluate the performance of aggregation schemes.

The results show that the performance of data aggregation is highly dependant on setup overhead, node density, sensing range, and the distance from sources to the sink. When the sources are close to the sink and the sensing range is short, tree-based aggregation, with long aggregation delay, achieves better performance in all metrics, except end-to-end delays. However, as the distance from sources to the sink and the sensing range increase, dynamic-cluster aggregation shows the best performance over other schemes, because a large number of generated packets are reduced by the local data aggregation.