Resource allocation policies for smart energy efficiency in data centers

Abstract

A data center must serve arriving requests in such way that the energy usage, reliability and quality of service performance should be balanced. This work is devoted to on-line resource allocation policies in data centers. We study a Markovian queueing system with controllable number of servers in order to minimize energy consumption and thus maximize the average revenue earned per unit time. An analytical model approach based on continuous-time Markov chain is proposed. The model is tested by simulation. Combining on-line measurement, prediction and adaptation, our techniques can dynamically determine the number of servers to handle the predicted workload. The policies comply with energy efficiency and service level agreements even under extreme workload fluctuations. © 2014 Springer International Publishing.