IrisNet : An Architecture for Internet-Scale Sensing

The time has come to consider wide-area architectures for pervasive sensing. Today’s low-cost PCs are deployed globally, connected to the Internet, and routinely have diverse powerful sensors attached to them, such as video cameras (webcams) and microphones. Even a PC’s high-speed network interface is a rich sensor, but one that senses the virtual environment of a LAN or the Internet, rather than the physical environment. Taken in the aggregate, the aforementioned hardware can be seen as an emerging global-scale sensor network. However, this sensor network lacks the architecture, algorithms, and software system needed to make it respond to users’ queries. We envision a global sensing system comprised of such rich, high-bit-rate sensor feeds—in which voluminous sensed data streams from vast collections of widely distributed sensors are available to users for querying as a single unit. To illustrate the type of wide-area sensing service we envision, imagine driving towards a destination in a busy metropolitan area. While stopped at a traffic light, you query a Parking Space Finder service using your PDA, by specifying your destination and criteria for desirable parking spaces (e.g., within two blocks of your destination, at least a fourhour meter). You get back directions to an available parking space satisfying your criteria. A half hour later, you return to your car and discover that it has been dented! By querying an Accident Witness service using your PDA, you retrieve images showing how your car was dented and by whom. These two example applications, a Parking Space Finder service and an Accident Witness service, support user queries over large collections of widely distributed video streams. In the IrisNet (Internet-scale Resource-Intensive Sensor Network Services) project at Intel Research Pittsburgh, we are designing and building an architecture and system that enable easy deployment of such wide-area sensing services over rich data feeds. To date, sensor network research has largely been defined by the design of algorithms and systems to cope with the severe resource constraints of battery-powered, tiny sensors which use wireless communication—slow CPUs, low-bitrate radios, and scarce energy. Such sensor networks are deployed over a single, contiguous communication domain. They use simple sensors that provide time series of single numerical measurements, such as temperature, pressure, light level, etc. Specialized hardware, operating systems, programming languages and database systems have been developed to accommodate this severely constrained environment [13, 16, 18]. In IrisNet, we seek to broaden the definition of “sensor network” to include the important, complementary class of wide-area sensor networks—Internet-connected, widelydispersed, PC-class nodes with powerful CPUs that can process rich sensor data sources. Indeed, such a wide-area sensing architecture is composable with existing low-power sensor networks: IrisNet is equally adept at integrating video streams from webcam-equipped PCs as at integrating sensed data from distinct, widely dispersed clouds of low-power wireless sensors. In this article, we offer a vision for wide-area sensing services; enumerate the technical challenges in achieving this vision; describe the IrisNet architecture, and how it meets these challenges; and report on present-day uses of the IrisNet system in several real wide-area sensing applications.