Data stream processing in sensor networks

Abstract

A sensor network consists of small computational devices that are able to communicate over wireless connection channels [44,40]. Each of these computational devices is equipped with sensing, processing and communication facilities. The sensing facility can sense physical values about the environment such as temperature, humidity and light. The processing part is able to do computation on the sensed values and/or other received values from the neighbors. The communication part is able to listen and send to other sensor nodes. The adoption of the sensor network technology will revolutionize our daily activities. Sensor networks will form a new world wide web that can read the physical world in real time [15]. Applications of sensor networks range from personal applications to scientific, industrial and business uses [11]. These applications will impact decision making at all levels, personally and professionally. Examples of these applications include habitat monitoring [31,39], animal control on a farm [10,38], traffic monitoring [19], underwater monitoring [41], fire detection [6], and smart homes [33]. Sensor networks generate data streams that need to be processed in real time for a wide range of applications. Therefore, there are two main aspects to this technology. The first is the scientific and technological aspect of sensor networks, which are discussed in Chap. 2. The other aspect is processing the resulting data streams, which is discussed in Chap. 3. This chapter is devoted to introducing data stream processing in sensor networks. The distinction between traditional data stream processing and sensory data stream processing is important because sensory data streams have their own features. Elnahrawy [14] distinguished sensor streaming from traditional streaming in the following way: • The sensor data are a sample of the entire population. On the other hand, traditional streaming data such as web logs and stock market data streams represent the entire population of the data. Basically the sensor data depends on the sampling rate. Some applications require higher rate than the others. For example, a sensor network that measures the temperature for meteorological purposes could sample the data every five minutes. However, if we deploy the same sensor network for physical experimentation in a scientific laboratory, the sampling rate could be several times per second. Thus, the sampling rate is an applicationdependent factor. The main theme in sensor networks is to convert continuous sensor readings to discrete ones. • The sensor data streams are considered noisy by comparison with other traditional streaming data. The state-of-the-art sensing equipment onboard sensor nodes requires data verification and validation. The environmental effect on the deployed sensor networks can also play a negative role on the sensed values. For example, web logs and web click streams are considered accurate values compared with data generated from sensor networks. Data quality assurance and cleaning are important to ensure the reliability of sensor networks. • The sizes of sensor data streams are usually less than traditional streaming data. This is valid for the current experimental deployment of wireless sensor networks. This will change in the near future with the expected large deployment of sensor networks to serve different applications. These features combined with new research challenges have introduced a new field of study, data stream processing in sensor networks. This field is concerned with handling and processing sensed data streams in wireless sensor networks. We can broadly classify the processing tasks in this area into data management and data analysis/mining. This chapter is organized as follows. Section 4.2 provides details about each class of data processing in sensor networks (data management and data mining). Section 4.3 highlights the research issues in data processing. Finally the chapter is summarized in Sect. 4.4. © 2007 Springer-Verlag Berlin Heidelberg.