Enforcing Quality of Service Using Decentralized Runtime Feedback Control

Abstract

The rapid growth of eCommerce increasingly means business revenues depend on providing good quality of service (QoS) for web site interactions. Traditionally system administrators have been responsible for tuning server configuration parameters, a process that is crucial to performance management, but time-consuming and skills-intensive. This paper describes an approach to automating performance management using a decentralized feedback control architecture. The feedback controller feeds back runtime service level measurements and forecasted workload characteristics, and automatically adjusts server tuning parameters to enforce the service level agreements from multiple classes of customers. Our studies using a multiple class discrete event simulator and a High Volume Web Sites (HVWS) simulator suggest that such a scheme can automatically adapt to changes in workload in order to reach the fixed Service Level Objectives (SLOs).