Reliability-Aware Speed Control Policy for Energy Reduction in Server Farms

Server farms are playing an important role in large-scale computing infrastructure today. However, the increasing energy consumption of server farms makes them expensive to operate. There are many technologies aiming at reducing the power consumption of server farms, for example, dynamic server speed scaling is an effective way to make the energy consumption proportional to the work load. But most of them have negative effect on system reliability due to increased transient failure rates. The repeated transition cycles increase the wear-and-tear of server components. In this paper, we control the process of speed scaling of a single server based on the dynamic workload using Markov Decision Process (MDP), and propose two reliability-aware algorithms for energy-efficient dynamic server speed control. Furthermore, we extend the model to multi-servers, it could provide speed control policies with performance, reliability and energy consumption concern. We evaluate the proposed models and policies via numerical experiments, which could show the effectiveness of our algorithms.