Data center selection based on neuro-fuzzy inference systems in cloud computing environments

Abstract

A high-quality service for applications in cloud computing environments is guaranteed by making efficient use of resources in data centers. Applications should be allocated to resources suitable for the load of data centers to achieve this. The complex and dynamic nature of the load prevents the proper selection of one of multiple data centers and fails to meet the demands of resources in applications. An incorrect data center selection seriously lowers resource utilization in the data center and accordingly deteriorates the quality of services for applications. This paper proposes a neuro-fuzzy inference-based prediction scheme to select one of multiple data centers in accordance with application workloads. This scheme is used to aggressively capture the time-varying load of data centers by learning and predicting the availability of resources therein. Therefore, it predicts not only the present load but also the future load of data centers in the process of determining a suitable data center. By an autonomic control for data center selection, our scheme can also provide load balancing between data centers. Moreover, we present performance evaluations with experiments based on Xen testbeds to demonstrate the effectiveness of our scheme. The experimental results show that our scheme is superior to other selection schemes with regard to the entire and changed loads of data centers.