A case for resource heterogeneity in large sensor networks

Sensor networks have traditionally consisted of nodes with the same amount of resources such as battery life and computational power. While such homogeneity has distinct advantages in terms of ease-of-fabrication, it has also been shown that in multi-hop ad-hoc scenarios homogeneity leads to large duty cycles, small end-to-end data throughput and poor deployment lifetimes. In this paper, we investigate sensor networks that have a single degree of heterogeneity, a random subset of the sensors, called accumulators, have more power and computational capability. To this end, we develop a decentralized, hierarchical clustering algorithm, called Hierarchical Clustering and Routing (HCR) algorithm. HCR exploits heterogeneity among sensor nodes to form a cluster hierarchy, with the objective of achieving better information throughput and improving network lifetime. A unique feature of HCR is its integration of routing with cluster formation and data delivery. Routing tables are constructed/updated in the course of cluster formation and data transmission, therefore incurring essentially no routing overhead. By exploiting geometrical features of hexagons, we show several desirable properties of HCR. In particular, we show via ns-2 simulations that HCR improves message overhead in cluster formation by 120-210%. We also show that heterogeneity achieves performance improvements of (i) upto 200% in terms of information throughput, (ii) power savings of upto 30% of the power spent in data transmission and (iii) a 2% density of accumulators is sufficient for most improvements.