Message Batching in Wireless Sensor Networks—A Perturbation Analysis Approach

We address the problem of batching messages generated at nodes of a sensor network for the purpose of reducing communication energy at the expense of added latency. We consider a time-based batching approach. We first develop baseline analytical models based on Markovian assumptions, derive conditions under which batching is profitable, and explicitly determine a batching time that optimizes a performance metric capturing the trade-off between communication energy and message latency. We then provide an on-line performance optimization method based on Smoothed Perturbation Analysis (SPA) for estimating the performance sensitivity with respect to the controllable batching time. We prove that the SPA gradient estimator is unbiased and combine it with a Stochastic Approximation (SA) algorithm for on-line optimization. Numerical results are provided for Poisson and Markov modulated Poisson message arrival processes and illustrate the effectiveness of the message batching scheme.