LPV Model and Its Application in Web Server Performance Control

Performance management is an important issue for Internet servers working in the unpredictable and highly dynamic environment. Feedback control is considered as a potential theoretical tool. However, due to the weakness in capturing the nonlinear natures of Internet servers, the classical linear control designs have been demonstrated ineffectiveness when handling varying workloads to provide sufficient performance guarantees. This paper attempts to apply linear-parameter-varying (LPV) approaches to modeling and controlling the Web server system for absolute connection delay guarantee. To achieve these goals, a LPV model, wherein the workload intensity is considered as the scheduling parameter, is identified experimentally to approximate the Web server system and then a LPV controller is designed for the LPV model. Model validation proves the LPV model is much more accurate than the LTI model with the same order. Closed-loop simulations under fluctuated workloads demonstrate that, the performance of the LPV controller overpasses the proportion-integral controller. By exploring the nature of dependence of server model on the workload intensity, the proposed LPV approach for absolute delay guarantee may find its applications in many other server performance control problems e.g., admission control, quality-of-service (QoS) control and power control.