Energy efficient data management

Abstract

Sensor networks have reached the stage of massive deployment for a wide range of applications such as ecosystem monitoring and traffic surveillance (Akyildiz et al. 2002; Szewczyk et al. 2004). A wireless sensor network typically consists of a base station (sometimes called sink, or gateway), and a set of sensor nodes (sometimes called sources, see Fig. 12.1). The sensor nodes are used to continuously capture environmental data such as temperature, humidity, lightness, etc. To send these data back to the base station and assist message routing, they are also equipped with radio transceivers that are capable of short-range transmission. The base station, as observed from the figure, serves as the gateway of the sensor network to exchange data and control messages with external applications to accomplish certain missions. As a typical example of the sensor network, suppose an external application wishes to know the average temperature of a certain area spanned by the sensor network. An explicit query (called an aggregation query) is then sent to the base station, which in turn forwards it to all the sensor nodes pertaining to this area. The temperature data (sometimes called readings or sensor values) are collected, averaged, and finally returned to the base station and the external application. capacity, they can assist the processing procedure in many ways. In what follows, we briefly introduce some basic notions that are frequently used by sensor network systems, and gradually elaborate them as we explore this topic. Snapshot query vs. continuous query. If the query involves a temporal parameter, which spans a limited or unlimited period of time, it is a continuous query. Such queries are usually used to monitor the physical world and its evolution over time. In contrast to a continuous query, a snapshot query only asks for the sensor data at a specified time instant. Query arrival rate vs. data capture (update) rate. For a particular sensor node, query arrival rate denotes the rate of incoming queries whose querying area overlaps the nodes coverage. This rate shows how popular this sensor node is, and it is often regarded as an external parameter (i.e., it cannot be controlled by the sensor network). The data capture (sometimes called "data update ") rate, on the other hand, denotes how often the sensor samples (sometimes called collects) the data. Different from query arrival rate, data capture rate is often tunable by the base station or the sensor itself. It is noteworthy that, these two rates are the two competing factors that affect the performance of the sensor network, similar to the consumer-producer relations. Push V.S. pull-based data dissemination. The collected data from a single sensor can be either pushed to the central server, i.e., the base station, or still kept locally at the sensor until some query explicitly pulls the data. Local V.S. centralized storage. From the storage point of view, push-based dissemination advocates centralized data storage while pull-based dissemination advocates local storage. It is also understood that centralized storage favors high query arrival rate and low data capture rate while local storage favors low query arrival rate and high data capture rate. Some recent work suggests hybrid pull-and-push-based dissemination, or in the storage terms, distributed sensor storage (Zhang et al. 2003). The idea is neither to send an overwhelming volume of raw sensor readings directly to the base station, nor to keep them at sensing locations, but rather, to organize and distribute them to appropriate locations throughout the sensor network so that these data can be quickly retrieved and queried later. Among the distributed storage proposals, data-centric storage is the most popular one (Ratnasamy et al. 2003). It names (or more formally, indexes) the data being stored for future searches. Both the base station and sensor nodes can access the data by their names across the entire sensor network, which explains why this storage scheme is called "data-centric ". This chapter mainly concerns the energy consumption issue during query processing in sensor networks. While the base station is supplied with unlimited power, the sensor nodes are usually battery-powered. In most cases, these batteries are inconvenient or even impossible to replace once they are deployed in the wild field. On the other hand, most sensor applications require continuous monitoring of data, which further makes energy efficiency a critical factor in the design of a good sensor network. Among all the sources that may consume sensor energy, wireless radio transmission is the most dominant source (Pottie et al. 2000). This is especially true in those networks where the sensor nodes fall far apart, since the power consumption for radio transmission is at least quadratic to the distance of transmission. The second major source that consumes significant energy is the sampling activities of the sensor nodes. Thus, a good design of energy-efficient sensor network should: (1) reduce unnecessary data transmission; (2) reduce the frequency of sensor reading. When a sensor node runs out of energy, its coverage is lost. In most scenarios, the mission of an application would not be able to continue if the total coverage loss over the entire area is remarkable. Formally, the network lifetime of a sensor network is the time duration until it fails to carry out the mission due to insufficient number of "alive " sensor nodes (Akyildiz et al. 2002). Energy- efficient data management is crucial for the sensor networks to accomplish the mission and accomplish it economically. However, this is a challenging task due to the following reasons. © Springer-Verlag Berlin Heidelberg 2007.