We present a new energy estimation model for sensor network intermediate gateway nodes (i.e. Crossbow Stargates). Such devices are battery powered and resource constrained and commonly employed as communication, processing, and gateway elements within sensor networks. Understanding and accurately estimating the energy behavior of such devices is key to conserving the battery life of the system.
Our approach considers the system as a whole and couples techniques that estimate energy consumption for a wide range of program activity, including computation, communication, and persistent storage access. We construct our model using empirical data that we collect via hardware performance monitors in the device and novel software performance monitors in the Linux operating system. By integrating information from the hardware, operating system, and program, we are able to accurately characterize the full system energy behavior of the device and its programs.
We extensively evaluate our model and compare its accuracy to that of an extant and similar approach to power estimation for the CPU and and memory subsystem of the Stargate. We find that this prior work, when applied directly to estimate whole system power consumption, is ineffective (introducing error rates of over 50%). Our model achieves an error rate of 3% for computationally bound tasks and of 11% for programs that employ both computation and communication.