A resource management framework for multi-tier service delivery in autonomic virtualized environments

Abstract

Large data centers usually host many different services on a shared computing infrastructure, for which on-demand resource management is necessary to maximize providers' revenues by meeting service quality targets at least operational cost. This paper presents a novel architecture of autonomic resource management framework based on virtualized service-oriented computing (SOC) environment. A non-linear continuous optimization problem is defined for adaptive resource allocation and a model-based approach is adopted to solve this problem. Different from traditional approaches, the analytic model we established provides probabilistic performance guarantees and considers non-steady-state behavior assisted by admission control. Results of prototype experiments demonstrate that the performance of multiple services has been greatly improved by taking advantage of fine-grained resource sharing, while incurring much lower resource usage cost. Also, differentiated service qualities could be provided to different client classes through our dynamic resource allocation scheme.