| Advances in radio, sensor, and VLSI technology will enable small and inexpensive sensor nodes capable of wireless communication and significant computation. Large-scale networks of such sensors may require novel data dissemination paradigms which are scalable, robust, and energy-efficient. In this dissertation, we design and evaluate directed diffusion, one such paradigm for distributed sensing applications in wireless sensor networks. Directed diffusion incorporates attribute-based naming, reinforcement-based adaptation, data-centric routing, and application-specific processing inside the network (e.g., data aggregation). By using attributes with external meaning at the lowest levels of communication, diffusion avoids multiple levels of name binding common to other approaches. Attribute-based naming in turn enables in-network processing with filters, supporting data aggregation, nested queries, and similar techniques that are critical to reducing network traffic and conserving energy. Given that, in wireless sensor networks, the communication cost is several orders of magnitude higher than the computation cost, directed diffusion can achieve significant energy savings with in-network data aggregation. We evaluate the energy efficiency of directed diffusion analytically and experimentally.;We also propose two instantiations of directed diffusion with different aggregation schemes: opportunistic aggregation and greedy aggregation. In the former approach, data is opportunistically aggregated at intermediate nodes on a low-latency tree. Our evaluation indicates that the opportunistic approach can achieve significant energy savings and can outperform idealized traditional schemes even with relatively unoptimized path selection. In the greedy approach, a greedy incremental tree is constructed to improve path sharing for more energy savings. Our result suggests that although greedy aggregation and opportunistic aggregation are roughly equivalent in low-density and medium-density networks, greedy aggregation can achieve significant energy savings in high-density networks without adversely impacting latency or robustness. |
