CAWSAC: Cost-Aware Workload Scheduling and Admission Control for Distributed Cloud Data Centers

• Published in: IEEE transactions on automation science and engineering 2016-04, Vol.13 (2), p.976-985
• Main Authors: Yuan, Haitao, Bi, Jing, Tan, Wei, Li, Bo Hu
• Format: Article
• Language: English
• Subjects: Admission control; Bandwidth; Cloud computing; cloud data centers (CDCs); Computer centers; Distributed databases; Distributed processing; Heuristic algorithms; Internet service providers; Scheduling; Servers; Time factors; traffic engineering; wide-area networks (WANs); workload scheduling; Workloads
• ISSN: 1545-5955; 1558-3783
• DOI: 10.1109/TASE.2015.2427234

Multiple heterogeneous applications concurrently run in distributed cloud data centers (CDCs) for better performance and lower cost. There is a highly challenging problem of how to minimize the total cost of a CDCs provider in a market where the bandwidth and energy cost show geographical diversity. To solve the problem, this paper first proposes a revenue-based workload admission control method to judiciously admit requests by considering factors including priority, revenue and the expected response time. Then, this paper presents a cost-aware workload scheduling method to jointly optimize the number of active servers in each CDC, and the selection of Internet service providers for the CDCs provider. Finally, trace-driven simulation results demonstrate that the proposed methods can greatly reduce the total cost and increase the throughput of the CDCs provider in comparison to existing methods. Note to Practitioners-A cloud provider deploys its applications in geographically distributed CDCs to improve stability and reliability. For cost and performance, each CDC provides services through multiple ISPs that deliver traffic between millions of users and the CDCs provider. The geographical diversity of the bandwidth and energy cost brings the CDCs provider a big challenge of how to minimize the bandwidth and energy cost of the CDCs provider. This paper first proposes a revenue-based workload admission control method to selectively admit requests. Then, this paper proposes a cost-aware workload scheduling method to allocate requests among multiple available Internet service providers connecting to distributed CDCs. The scheduling strategy can intelligently dispatch requests, and achieve lower cost and higher throughput for the CDCs provider.