A Strategy for Queuing Theory-Based Performance and Energy Management in Heterogeneous Data Centers

Abstract

With the increasing popularity of Internet services and cloud computing, both energy efficiency and performance guarantees are major concerns for data center operators. However, most of the previous works only address this problem in homogeneous data center environments. This paper investigates how to guarantee performance requirement in terms of waiting time so as to minimize power consumption among multiple types of servers with different processing capabilities in a data center. We address the problem of minimizing the energy consumption of a data center by means of dynamic management policies that switch on/off a certain quantity of servers necessary under given performance requirements in terms of waiting time. We develop a mathematical model using queuing theory to determine the number of servers that should be used i.e., powered on. Experiment results show that our scheme can reduce power consumption compared with the non-group division strategy. Our scheme provides an analytical approach for studying the power-performance tradeoff in the heterogeneous data center.