1: \begin{abstract}

2: In this paper, we use a Markov decision process to find optimal asynchronous

3: policy of an energy-efficient data center with two groups of heterogeneous

4: servers, a finite buffer, and a fast setup process at sleep state. Servers in

5: Group 1 always work. Servers in Group 2 may either work or sleep, and a fast

6: setup process occurs when server's states are changed from sleep to work. In

7: such a data center, an asynchronous dynamic policy is designed as two

8: sub-policies: The setup policy and the sleep policy, which determine the

9: switch rule between the work and sleep states for the servers in Group 2. To

10: analyze the optimal asynchronous dynamic policy, we apply the Markov decision

11: process to establish a policy-based Poisson equation, which provides expression

12: for the unique solution of the performance potential by means of the

13: RG-factorization. Based on this, we can characterize the monotonicity and

14: optimality of the long-run average profit of the data center with respect to

15: the asynchronous dynamic policy under different service prices. Furthermore,

16: we prove that the bang-bang control is always optimal for this optimization

17: problem, and supports a threshold-type dynamic control in the energy-efficient

18: data center. We hope that the methodology and results derived in this paper

19: can shed light to the study of more general energy-efficient data centers.

20: 

21: \textbf{Keywords:} Energy-efficient data center; Asynchronous dynamic policy;

22: Markov decision process; Sensitivity-based optimization; RG-factorization.

23: \end{abstract}