Measurement and monitoring in wireless sensor networks

• Authors: Ramesh Govindan;Yonggang Zhao
• Venue: Measurement and monitoring in wireless sensor networks
• Year: 2004

Wireless sensor network is one of the dominant technology trends in the coming decades. These networks will consist of a large collection of sensors and/or actuators, collaboratively providing robust service with constrained energy and bandwidth resources. Systematic measurement of network performance is one important component in the iterative process of design and evaluation. A monitoring infrastructure that indicates failures, resource depletion, and other abnormalities, is a key component of any operational sensor network system. Sensor networks present unique challenges for measurement and monitoring: its large-scale deployment, complex collaboration algorithms, resource constraints, and high device-to-user ratios. We argue that heavy environmental dependence, complex collaborative behavior and large network dynamic range are three primary challenges for sensor network measurement and monitoring. Several principles in measurement are presented to address each of these issues in sensor networks. We also propose a new architecture to collect network status in different spatial or temporal details. We identify several design principles to ensure energy-efficiency, scalability and robustness: in-network aggregation to push data processing into the network, vertical integration to reduce unnecessary messaging overhead, and localized signaling to increase tolerance to routing failures. We describe in detail a systematic measurement on packet delivery performance. The environmental dependence is carefully accounted by experimentation in different environments. Additionally, the dynamic range of packet delivery are measured by fine-grain long-term experiments at different network layers: physical layer and MAC layer. We show that how those experiments are carefully designed to cover a wide range of parameters and metrics. We present the design and evaluation of two major components in our monitoring architecture: sensor network scans and digests. Escans that approximately depicts the remaining energy distribution within state. Digests that can continuously provide updates of key system-wide metrics, by efficiently computing aggregate function ( e.g., sum, average, count) of network properties with only local information exchange. Finally, we describe our experience in applying our approaches in the testbeds and the implications on design and implementation of sensor networks.