Application-specific modelling of information routing in wireless sensor networks

Sensor network applications have a diverse set of requirements - some involve extraction of sensor data to a single point, others exploit sensor-to-sensor communication; some employ long-lasting data streams while connections in others are mainly ephemeral. Different variants of the directed diffusion routing protocol - pull-based, push-based and hybrid rendezvous-based - have been developed, along with in-network processing and geographic routing techniques. We mathematically model and analyze the performance of these routing techniques across a range of application scenarios (with varying numbers of nodes, sources, sinks, data settings etc.). Besides quantifying the conditions under which the different routing algorithms outperform each other, we obtain a number of useful design insights. Our analysis shows that algorithms mismatched to applications can result in drastically poor performance; demonstrates the desirability of reducing flooded interest and exploratory messages when data aggregation is used; and suggests that it may be difficult to implement efficient hybrid schemes because their performance is very sensitive to the optimal placement of rendezvous points.