Conceptually, rate control algorithms have two key objectives: first, they regulate flows, that have arrival rates outside the feasible rate region of the network, in order to operate them within the rate region. Second, they control the point, within the rate region, at which the network operates in order to optimize a specific global network utility function. Specifically for wireless sensor networks, flows are generally short-lived, and the network utility function that the rate control algorithm needs to optimize for is application-dependent. Given short flow life-spans, rate control algorithms for these networks thus need to have fast convergence times. Further, given the diversity of applications targeted for sensor networks, rate control stacks for these networks need to be modular enough to optimize for a multitude of network utility functions. Current state-of-the-art protocols in this space provide monolithic stacks based on AIMD mechanisms that do not sufficiently satisfy these needs.

We have designed two rate control protocols to address these concerns. The first protocol, is the Wireless Rate Control Protocol (WRCP), which uses explicit knowledge of the available capacity to achieve lexicographic max-min fairness, over a sensor network collection tree. At the core of WRCP is a novel interference model that simplifies the calculation of achievable capacity in a multi-hop CSMA-based wireless sensor network. Explicit knowledge of the achievable capacity allows WRCP to exhibit excellent convergence times when compared to state of the art AIMD mechanisms. The second protocol is the Backpressure based Rate Control Protocol (BRCP). BRCP has been designed using stochastic optimization techniques that present functions which map the current queue size at a source to admissible flow rates. The mapping function depends on the network utility function that the protocol is attempting to optimize. Since mapping functions can be generated for any concave utility function, BRCP presents a modular architecture that allows for a wide range of optimization goals. Both protocols have been implemented on the TinyOS-2.x stack, and have been evaluated over the USC Tutornet testbed .