Sleeping policy cost analysis for sensor nodes collecting heterogeneous data

Abstract

Sleeping policies are widely used in many contexts to reduce energy consumption in sensor nodes, ad-hoc terminals or mobile cell phones. Although there is extensive literature on energy saving in sensor networks, little has been done with respect to comparing additional costs, such as packet drop probability and buffer occupancies of sleeping versus non-sleeping policies, especially for multiple classes of data. In this paper, we employ discrete time vacation modeling results from queueing theory to obtain the quantities of interest, i.e., blocking probabilities, and queue lengths incurred by sleeping policies for a sensor node with a finite buffer.We propose a 2-class Geo/G/1/K vacation model, where two classes of data, each with its own buffer, are collected by a sensor node following a sleeping policy. At any time instant, only one packet can be transferred and the transmission order is selected by a scheduling function. The sleeping, transfer and selection processes have general distributions. To the best of our knowledge, this is the first work on modeling and analyzing the performance characteristics of a sensor node with a sleeping policy, collecting heterogeneous data. First, we present a Markov renewal process embedded at the time instant of transfer completion and then obtain the marginal occupancy distributions at an arbitrary time instant. Simulation results validate our theoretical derivation and demonstrate how the results can be applied to understand the behavior of sensor nodes with heterogeneous data.