Cost-adaptive load sharing for geographically distributed data center network

Abstract

A scalable framework that uses a cost-adaptive load sharing among Internet data centers (IDC) deployed over a large scale geographic area is introduced in this study. The key idea is to efficiently allocate load to the IDCs, in order to reduce the cost of cloud computing. Data centers are designed and built based on their expected peak load. Thus, reducing the expected peak load can significantly reduce their capital expenses (e.g., their server acquisition cost). Previous studies addressing the issue of reducing the cost of cloud computing focused on reducing its electricity cost, while ignoring the capital cost, which is significantly higher (Hamilton in ACM SIGCOMM CCR 29(1):68–73, 2009; https://www.forbes.com/sites/.../09/.../cost-wars-data-center-vs-public-cloud/ ; https://ongoingoperations.com/data-center-pricing-credit-unions/ ; http://perspectives.mvdirona.com/2010/09/overall-data-center-cost/ ). Thus, these methods are not scalable, and their potential contribution and possible implementation are not clear. This study shows that minimizing the IDC electricity cost without conducting load balancing among the IDCs may eventually increase the total cost associated with global cloud computing, due to the potential increment in the IDC peak load. This study shows that balancing load, electricity cost, and networking cost, subject to service delay constraints, the entire cost of cloud computing (CapEx plus OpEx) can be further reduced, in comparison with previous studies.