Energy Consumption Characteriation of Heterogeneous Servers

Energy and cooling cost is becoming the main cost of data centers. Many studies focused on how to schedule job and migrate Virtual Machine between servers to save energy. An accurate energy consumption model is the basic of energy management. Most past studies show that energy consumption has linear relation with resource utilization. We found that different servers have different energy consumption characters even with same CPU and similar workloads. Most past models were only validated on several servers. These models did not reflect the characters of different kinds of servers in a heterogeneous data center. In this paper, we verified the accuracy of the linear model on 392 servers. The data come from the results of SPECPower_ssj2008 benchmark. The benchmark was released in 2007. There are 392 servers from 26 vendors were tested in the past six years. These servers represent most common used servers in heterogeneous data centers. We use several methods to validate the accuracy of the linear model. The results show the model does not work well for all kinds of servers. We also analyze the trend of energy consumption of last six years.