Identifying Energy Waste through Dense Power Sensing and Utilization Monitoring

PowerNet is a hybrid sensor network for monitoring the power and utilization of computing systems in a large academic building. PowerNet comprises approximately 140 single-plug wired and wireless hardware power meters and 23 software sensors that monitor PCs, laptops, network switches, servers, LCD screens, and other office equipment. PowerNet has been operational for 14 months, and the wireless meters for three months. This dense, long-term monitoring allows us to extrapolate the energy consumption breakdown of the whole building. Using our measurements together with device inventory we find that approximately 56% of the total building energy budget goes toward computing systems, at a cost of ≈ $22,000 per month. PowerNet’s measurements of CPU activity and network traffic reveal that a large fraction of this power is wasted and shows where there are savings opportunities. In addition to these sensor data results, we present our experiences designing, deploying, and maintaining PowerNet. We include a longterm characterization of CTP, the standard TinyOS collection protocol. The paper concludes with a discussion of possible alternatives to computing system design that can save energy while satisfying user workloads.