Rbuff: Improving congestion in sensor networks under event-driven and burst data traffic

Abstract

Advancements in sensor and MEMS technology have enabled high resolution, high data-rate, and complex sensors which enhance the application domain of sensor networks. In addition, these sensors enable sensor networks to capture high quality data with more precision. While increased storage capacities on sensor nodes have previously enabled sensor networks to store and forward data leisurely, many emerging sensor network applications, such as seismic monitoring, real-time object localization and tracking, or pervasive health monitoring, require real-time reporting of this high resolution, event-driven data. The existing communication and radio stack in sensor network operating systems were designed for simple packet handling; however they fail under high data-rate and burst traffic. In this work, we propose a modified communication stack which includes a receive buffer (RBuff) to handle burst traffic more efficiently, reducing traffic congestion. We present a theoretical analysis on the optimal buffer size based on the properties of the expected burst traffic within the network. In addition, we address the dual scenario; we present analysis to determine the maximum burst size and wait time given a limited fixed buffer size. Experimental analysis on single-hop, multi-hop forwarding trees, and random network deployments demonstrates a 50% increased packet reception rate under burst traffic of the optimally sized Rbuff over the existing single packet slot within the Contiki operating system. Additionally, we show that a fixed buffer implementation with pre-determined burst sizes and wait times also provide better results than the single buffer implementation. We demonstrate how a modest buffered approach improves packet reception in event and burst traffic scenarios and aids in reducing overall network energy consumption by reducing collisions.