Web Traffic Modeling and Server Performance Analysis

In this paper we present a study of the traffic patterns for a dynamic and heavily-accessed Web server environment, and the impact of such traffic patterns on Web server performance. Using the data from the official Web site during the 1998 Winter Olympic Games in Nagano, Japan, we develop traffic models to represent the process of user requests that are the input to the geographically-distributed Web server systems. Our detailed analysis of the traffic data illustrates traffic patterns that exhibit both light-tailed and heavy-tailed behaviors, some of which appear to be quite different from the Web server systems modeled as a general single-server queue and analyze the waiting-time process, which models the latency encountered by user requests. Our mathematical analysis characterizes the tail behavior of the waiting-time distribution, i.e., the probability that the latency exceeds a certain level, which is a key measure of quality of service (QoS).