Percentile performance estimation of unreliable IaaS clouds and their cost-optimal capacity decision

Abstract

Through Internet, a cloud computing system provides shared resources, data, and information to users or tenant users in an on-demand and pay-as-you-go styles. It delivers large-scale utility computing services to a wide range of consumers. To ensure that their provisioned service is acceptable, cloud providers must exploit techniques and mechanisms that meet the service-level-agreement (SLA) performance commitment to their clients. Thus, performance issues of cloud infrastructures have been receiving considerable attention by both researchers and practitioners as a prominent activity for improving service quality. This paper presents an analytical approach to percentile-based performance analysis of unreliable infrastructure-as-a-service clouds. The proposed analytical model is capable of calculating percentiles of the request response time under variable load intensities, fault frequencies, multiplexing abilities, and instantiation processing time. A case study based on a real-world cloud is carried out to prove the correctness of the proposed theoretical model. To achieve optimal performance-cost tradeoff, we formulate the performance model into an optimal capacity decision problem for cost minimization subjected to the constraints of request rejection and SLA violation rates. We show that the optimization problem can be numerically solved through a simulated-annealing method.