INRIA MISTRAL Group Position Paper

by Zhen LIU

The MISTRAL group at INRIA has been working on the performance issues of the Web since a couple of years. Our past and on-going work are concerned with characterization of Web traffic, caching policies and caching architectures, and modeling and performance evaluation of Web servers and Web applications. Our interest in attending the W3C Workshop on Web Characterization is to exchange information and interact with people in the community of Web traffic analysis. We will present our recent work on the Web traffic modeling and performance comparison between HTTP1.0 and HTTP1.1. We would like to contribute to the traffic modeling and traffic generation as well as the experimentation of the HTTP-NG within the framework of WCG-II.


Web traffic modeling and performance comparison between HTTP1.0 and HTTP1.1

Abstract

In this work we propose a hybrid model for describing the Web traffic at request level. We concentrate on the HTTP requests to the same Web server and we are interested in the typical behavior of the clients' hypertext navigation. The mathematical model is defined by a stochastic marked point process which describes when clients arrive and how they browse the server. We have developed a benchmark tool of Web servers: WAGON (Web trAffic GeneratOr and beNchmark). It comprises a generator of Web traffic, a robot which sends and analyzes requests and a monitoring tool. It can generate different types of traffic requests, which, in turn, can be sent out to the server from different machines with different (simulated) delays and bandwidths. Using such a tool we have carried out experiments and compared the performance between HTTP1.0 and HTTP1.1 under different configurations. Our results show in particular that HTTP1.1 almost always outperform HTTP1.0, and this is true even if the latter uses more parallel sessions and slightly larger bandwidth. The difference becomes more significant when the pipeline mechanism is used in HTTP1.1 and when the propagation delay is large.