Talk
"Data-Driven Query Processing in Sensor Networks"
Adam Silberstein, Duke University
Wednesday, October 25, 2006 at 3:00 P.M.
Lubrano Conference Room (4th Floor CIT)
Wireless sensor networks have become a popular field of research, spawning new applications and new fundamental research challenges. A sensor network may consist of a number of battery-powered nodes, each equipped with sensing devices that generate measurements, and a radio for communicating to other nodes over short distances. The network is rooted at a more powerful base station, with which the user interacts. For many applications, a sensor network can be seen a distributed data source supporting continuous queries under severe resource constraints, giving rise to many challenges not present in traditional query processing.
Because of limited energy on battery-powered nodes, a primary goal is minimizing the amount of radio transmission, often the dominant energy consumer in sensor networks. In this talk I will introduce a data-driven approach to sensor data processing, which exploits models of sensor node behavior for optimization, but ultimately considers the data to be the ground truth. A primary technique in this approach is suppression, which installs constraints among the network nodes, such that the nodes only transmit when constraints are broken. I will discuss in detail work on continuous max aggregation, presenting a series of algorithms that exploit suppression in increasingly sophisticated ways. Finally, I will overview two interesting challenges that arise from suppression and in-network constraints: coping with message failures, and managing the competing goals of the query application and the underlying network routing layer.
Host: Ugur Cetintemel