Predicting the QoS of an electronic commerce server

Abstract

This paper presents a case study on Quality of Service (QoS) measures and Service Level Agreements (SLA) for an electronic commerce server. Electronic commerce systems typically rely on a combination of an HTTP server and a database server that may be integrated with other enterprise information resources. Some interactions with these systems cause requests for static HTML pages. Others cause significant amounts of database processing. Response time percentiles are well-accepted measures of QoS for such requests. In this paper we measure the behavior of an electronic commerce server under several controlled loads and study response time measures for several workload abstractions. Response time measures are captured for individual URLs, groups of functionally related URLs, and for sequences of URLs. We consider the utility of these workload abstractions for providing SLA. We also show that empirical evidence of server behavior in conjunction with analytic modeling techniques may be useful to predict the 90-percentile of response times for sequence based workload classes. The model predictions could be used to support realtime call admission algorithms.