On the modeling of WWW request arrivals

In this paper, access logs of two Web servers are analysed. The objectives of the analysis are to characterize the arrival process of a Web server, and to build an arrival model based on the characterization. Our analysis finds that the arrival process is actually formed by superimposing different users' request sessions which have an exponentially distributed inter-arrival time. Each session is also found to consist of one or more HTML document requests which is geometrically distributed, while the think time has a Weibull distribution. A number of requests are generated based on an HTML document which is found to be geometrically distributed, with their inter-arrival time having a Weibull distribution. Based on these characterizations a model on WWW request arrivals can be built easily. The validity of this model is verified by comparing the arrivals generated by the model to the two access logs. Our results show that the inter-arrival distribution of the generated arrivals match very well with those of the access logs.