1: \begin{abstract}

2: %\boldmath

3: Efficient virtual machine (VM) management can dramatically reduce energy

4: consumption in data centers. Existing VM management algorithms fall into

5: two categories based on whether the VMs' resource demands are assumed to

6: be static or dynamic. The former category fails to maximize the resource

7: utilization as they cannot adapt to the dynamic nature of VMs' resource

8: demands. Most approaches in the latter category are heuristical and lack

9: theoretical performance guarantees.

10: In this work, we formulate dynamic VM

11: management as a large-scale Markov Decision Process (MDP) problem and derive an optimal solution. Our

12: analysis of real-world data traces supports our choice of the

13: modeling approach. However, solving the large-scale MDP problem suffers

14: from the curse of dimensionality. Therefore, we further exploit the special structure of the problem and propose an

15: approximate MDP-based dynamic VM management method, called MadVM. We

16: prove the convergence of MadVM and analyze the bound of its approximation

17: error. Moreover, MadVM can be implemented in a distributed system, which

18: should suit the needs of real data centers. Extensive

19: simulations based on two real-world workload traces show that MadVM

20: achieves significant performance gains over two existing baseline

21: approaches in power consumption, resource shortage and the number of VM

22: migrations. Specifically, the more intensely the resource demands fluctuate, the more MadVM outperforms.

23: \end{abstract}

24: