Year

2004

Degree Name

Doctor of Philosophy

Department

School of Electrical, Computer and Telecommunications Engineering - Faculty of Informatics

Abstract

This thesis proposes a new model for the marginal distribution of HTTP request rate. The model applies to aggregate network traffic generated by a population of users accessing the Web on the Internet. The new model is relatively simple and allows for both the accurate estimation of peak HTTP request rate and the development of two new rules of thumb concerning HTTP request rate. Previous models of HTTP request rate have generally been single user models of a form that are both complex to transform into a model of aggregate traffic and apply to the estimation of peak aggregate HTTP request rate. One comparable model of aggregate HTTP traffic models HTTP request inter-arrival time rather than HTTP request rate and is shown to over estimate peak HTTP request rate. There are few existing rules of thumb concerning HTTP request rate. The two rules proposed here are the first for the estimation of either standard deviation or peak HTTP request rate at the second time scale. The new model for the marginal distribution of aggregate per second HTTP request rate is based on the P�lya-Aeppli probability distribution. The selection of the P�lya-Aeppli distribution can be justified from observed distributions of HTTP request rate of individual Web users and the number of active users per second in a population of Web users. The results are based on the analysis of five independent traces of Web traffic. One trace, collected by the candidate, is of per-user Web traffic generated in a postgraduate research laboratory at the University of Wollongong (UOW) between 1994 and 1997. The other four traces are large independent traces of aggregate Web traffic collected between 1996 and 2002.

02Whole.pdf (3091 kB)

Share

COinS