Make a Sensor Net Work

The miniaturization of computing, communication and sensor devices enables researchers to deploy wireless sensor networks, which extend computation capabilities into the physical world. Recent research on wireless sensor networks has been very productive, with more than a dozen platforms developed and a wide range of application areas explored, including environment monitoring, border control, wildlife tracking, sniper localization, structural defect detection, and large-scale surveillance. On the other hand, there are still two important questions that have not been completely answered for the future of wireless sensor network technology: First, with inexpensive hardware and a stringent energy budget, can wireless sensor networks provide functionality that significantly distinguishes it from other technology? Second, is the current wireless sensor network technology realistic enough to be used in the real world and expect a market success? Our research tries to answer these two questions. We have developed VigilNet, a large-scale surveillance network that classifies four types of targets: people, people with ferrous objects, small vehicles, and large vehicles. It is likely the largest wireless sensor network software that has been developed. More importantly, it includes functionality that is beyond traditional sensor technology. Second, we have designed a new OS kernel, t-kernel,that significantly enhances wireless sensor network platforms' functionality and facilitates development on such platforms. In contrast with traditional OS's for embedded systems, t-kernel supports virtual memory without address-translation hardware, protects the OS from application code without a separate address space, and provides reemptive priority scheduling without privileged instructions. Although t-kernel slows down the execution speed of applications, it transforms the "cheap" and unreliable hardware platform into a realistic platform that people can control, maintain, and program.