Skip to Main content Skip to Navigation

Etude des performances d'un serveur web et d'un réseau local sans fil utilisant les techniques neuronales

Abstract : The popularity of web service raises a lot of problems related to its performance. The use of "intelligent" mechanisms helps to overcome the difficulty linked to more traditional methods and takes into account the whole complexity due to the introduction of parameters influencing the quality of service of a web server. This is the context of the research in this thesis. We are interested in the problem of web server performance evaluation and in the access mechanism for infrastructure based wireless network. Instead of using the traditional methods to modelize their performance, we propose the use of the learning and generalisation capacities of the neural networks to learn these performances from the data obtained from experimentation and simulation. First, we study the effects of the different influencing parameters of a Web server on its performance metrics. The results help to give ideas to system and network administrators on the ways to adjust the server parameters in order to improve its performance. We also propose a model based on a simple queue representing the web server architecture (processor, memory and disk) by using the iterative technique based on the MVA (Mean Value Analysis) equations. Next, this thesis proposes a new approach based on the use of neural networks to modelize the web server performance by taking its optimisation parameters into account. These models were then implemented on a real web server, Apache with FreeBSD operating systems in order to put in place an overload control mechanism for the server. This control strategy allows to avoid overloading the server. A mixed control mechanism based on the combination of two controllers, an open-loop based neural and a closed-loop, Proportional Integral, was elaborated. These two controllers help together to improve in a considerable way the admission control mechanism of the web server. Finally, our interest is on the performance modelling of the access mechanism for infrastructure based wireless network. Two models, mathematical and neuronal, are used for estimating the maximum throughput observed on each wireless station as a function of the packet length, the transmission rate and the number of stations for the UDP and TCP protocols .The results obtained help the user to know the capacity of a hot spot of its wireless network.
Complete list of metadatas

Cited literature [51 references]  Display  Hide  Download
Contributor : Camille Meyer <>
Submitted on : Monday, August 27, 2012 - 10:55:48 AM
Last modification on : Monday, January 20, 2020 - 12:12:05 PM
Long-term archiving on: : Friday, December 16, 2016 - 8:08:55 AM


  • HAL Id : tel-00725472, version 1


Fontaine Rafamantanantsoa. Etude des performances d'un serveur web et d'un réseau local sans fil utilisant les techniques neuronales. Réseau de neurones [cs.NE]. Université Blaise Pascal - Clermont-Ferrand II, 2009. Français. ⟨NNT : 2009CLF21935⟩. ⟨tel-00725472⟩



Record views


Files downloads