Many of the proposed mechanisms aiming to achieve energy-aware adaptations in server environments rely on the existence of models that estimate the power consumption of the server as well as its individual components. Most existing or proposed models employ performance (hardware) monitoring counters and the CPU utilization to estimate power consumption, but they do not take into account the statistics of the workload the server processes. In this paper we propose a lightweight probabilistic model that can be used to estimate the power consumption of the CPU, the network interface card (NIC), and the server as a whole. We tested the model's accuracy by executing custom-made benchmarks as well as standard benchmarks on two heterogeneous server platforms. The estimation error associated with our model is less than 1% for the custom-made benchmark whereas it is less than 12% for the standard benchmark.
展开▼