Self-Diagnosis and Self-Regulation through Performance Monitoring and Tuning

Abstract

Although often relegated behind other more critical qualities of service, performance has a number of characteristics that make it an ideal target for monitoring and assurance. First, there are a number of metrics easily accessible that are indicators of a decrease in performance. Second, a lack of availability is often preceded by a decrease in performance. Finally, many functional problems emerge only when resources are stressed and performance is low. These characteristics are the motivation we use to develop an approach by which web-service systems are able to detect problems before they occur, and take appropriate corrective actions without human intervention, or trigger alarm and notification processes to request human intervention.