Microservers---32bit nodes with megabytes of RAM, sophisticated peripherals and high-bandwidth radios---are becoming increasingly important in wireless sensor network applications, since their capabilities allow for substantially more complex processing and sensing. One of the biggest drawbacks of microserver nodes compared to mote-class devices is their increased energy consumption. Until recently, most of the research effort in reducing energy consumption has focused on hardware solutions.

Improvements such as energy efficient CPUs with voltage scaling, usage of low-power NAND flashes and multiple heterogeneous radio interfaces have resulted in significant energy savings in newer generations of microservers. On the other hand, not much effort has been directed on the software side of things; microservers typically use generic purpose OSes that are not specifically designed with energy optimization in mind.

In this talk we will present some initial steps towards "energy-aware operating systems". Our main goal is to augment and enhance existing OSes so that they can provide energy-efficient operation on the software level (e.g. network interface selection, energy-aware scheduling) as well as on the hardware level (e.g. selecting the optimal CPU/RAM/networking speed for a particular operation).

We will present a tool, called etop that displays energy consumption on a per-process, per-subsystem basis in real-time. We will also show initial results on the real-time energy behavior of a node under different operating conditions such as network congestion and under different workloads (CPU intensive applications, I/O intensive applications etc), as well as some simple optimizations to reduce energy consumption.