An LPV Approach to Performance Control of Web Servers Under Self-Similar Workloads

Applying control-theoretic approaches to capacity provisioning of web servers is gaining popularity. This paper presents a novel approach that combines Linear Parameter Varying (LPV) modeling and control design, with workload characterization using stochastic α-stable-model envelopes. We parameterize a control-oriented Web server model and the resulting controller using workload distribution parameters. By further incorporating the α-stable modeling into the LPV control approach, the presented solutions not only allow system to adapt to workload changes, but also show great promise in handling self-similar workloads. The proposed method is applied to a CPU allocation problem for web servers, which performs Dynamic Voltage Scaling (DVS) to achieve response time guarantee. Simulations using real web-server traces are conducted to show the strength of the proposed approach. I. INTRODUCTION