Data Center Selection Based on Reinforcement Learning

Abstract

As demand for cloud services continues to increase, major cloud service providers have built many data centers around the world or in certain regions to ensure the quality of services. However, optimal data center selection from the distributed data centers of a cloud service provider or a cloud service alliance remains a problem, in part because of the cost of leasing data center resources, which includes the costs of network resources and computing resources. Whereas the cost of computing resources is easy to determine, the cost of network resources is relatively difficult to determine because it depends on communication paths. In this study, the distributed data centers were abstracted into an incomplete undirected graph. Furthermore, the cost of computing resources and network resources were considered together. The mathematical model of the user’s cost was established. To obtain the optimal solution, we proposed a data center selection algorithm based reinforcement learning. The experiment results showed that the algorithm effectively solved the problem.