Model Identification for Performance Management of Distributed Enterprise Systems

Model-based techniques have been explored recently by researchers aiming to develop effective autonomic management techniques for multi-tier enterprise systems under uncertain and dynamic operating conditions. The general aim is to minimize operational costs while maximizing a multidimensional QoS metric that typically includes service related factors such as response time, throughput, and reliability. In this paper we present a systematic approach to develop accurate models for distributed multi-tier enterprise systems. The proposed approach first identifies the system parameters through extensive experimentation and then defines the relationship among these parameters and identifies the underlying model structure of the system using regression methods and bayesian estimation techniques. The developed model can then be used in various model-based autonomic management structures used for performance control, fault adaptation, and security management. In this techincal report, we demonstrate the effectiveness of our approach by using a predictive controller to reduce power consumption of a multi-tier enterprise system and still achieve a good response time.