Typically a wireless LAN infrastructure is designed and installed by Networking professionals. These individuals are extremely familiar with wired networks, but are often unfamiliar with wireless networks. Thus, Wireless LAN installations are currently handicapped by the lack of an accurate, performance prediction model that is intuitive for use by non-wireless professionals. To provide a solution to this problem, this thesis presents a method of predicting the expected wireless LAN throughput using a site-specific model of an indoor environment. In order to develop this throughput prediction model, two wireless LAN throughput measurement products, LANFielder and SiteSpy, were created. These two products, which are patent pending, allow site-specific network performance measurements to be made. These two software packages were used to conduct an extensive measurement campaign to evaluate the performance of two IEEE 802.11b access points (APs) under ideal, multiuser, and interference scenarios. The data from this measurement campaign was then used to create empirically based throughput prediction models. The resulting models were first developed using RSSI measurements and then confirmed using predicted signal strength parameters. / Master of Science

L'évolution des télécommunications amené aujourd'hui à un foisonnement des appareils connectés et une massification des services multimédias. Face à cette demande accrue de service, les opérateurs ont besoin d'adapter le fonctionnement de leurs réseaux, afin de continuer à garantir un certain niveau de qualité d'expérience à leurs utilisateurs. Pour ce faire, les réseaux d'opérateur tendent vers un fonctionnement plus cognitif voire autonomique. Il s'agit de doter les réseaux de moyens d'exploiter toutes les informations ou données à leur disposition, les aidant à prendre eux-mêmes les meilleures décisions sur leurs services et leur fonctionnement, voire s'autogérer. Il s'agit donc d'introduire de l'intelligence artificielle dans les réseaux. Cela nécessite la mise en place de moyens d'exploiter les données, d'effectuer surelles de l'apprentissage automatique de modèles généralisables, apportant l’information qui permet d'optimiser les décisions. L'ensemble de ces moyens constituent aujourd'hui une discipline scientifique appelée science des données. Cette thèse s'insère dans une volonté globale de montrer l'intérêt de l'introduction de la science des données dans différents processus d'exploitation des réseaux. Elle comporte deux contributions algorithmiques correspondant à des cas d'utilisation de la science des données pour les réseaux d'opérateur, et deux contributions logicielles, visant à faciliter, d'une part l'analyse, et d'autre part le déploiement des algorithmes issus de la science des données. Les résultats concluants de ces différents travaux ont démontré l'intérêt et la faisabilité de l'utilisation de la science des données pour l'exploitation des réseaux d'opérateur. Ces résultats ont aussi fait l'objet de plusieurs utilisations par des projets connexes. / The evolution of telecommunications has led today to a proliferation of connected devices and a massification of multimedia services. Faced with this increased demand for service, operators need to adapt the operation of their networks, in order to continue to guarantee a certain level of quality of experience to their users. To do this, operator networks tend towards a more cognitive or autonomic functioning. It is about giving the networks the means to exploit all the information or data at their disposal, helping them to make the best decisions about their services and operations,and even self-manage. It is therefore a questionof introducing artificial intelligence into networks. This requires setting up means to exploit the data, to carry out on them the automatic learning of generalizable models, providing information that can optimize decisions. All these means today constitute a scientific discipline called data science. This thesis fits into a global desire to show the interest of the introduction of data science in different network operating processes. It inlcudes two algorithmic contributions corresponding to use cases of data science for the operator networks, and two software contributions, aiming to facilitate,on the one hand, the analysis, and on the other hand the deployment of the algorithms produced through data science. The conclusive results of these various studies have demonstrated the interest and the feasibility of using data science for the exploitation of operator networks. These results have also been used by related projects.

Performance evaluation continues to play an important role in network research. Two types of research efforts related to network performance evaluation are particularly noteworthy: (1) using performance evaluation to understand specific problems and to design better solutions, and (2) designing efficient performance evaluation methodologies. This thesis addresses several performance evaluation challenges, encompassing both categories of effort listed above, in building high-performance TCP-based network services in the context of overlay routing and peer-to-peer systems. With respect to the first type of research effort, this thesis addresses two issues related to the design of TCP-based network services: 1. Prediction of large transfer TCP throughput: Predicting the TCP throughput attainable on given paths is used for applications such as route selection in overlay routing. Based on a systematic measurement study, we evaluate the accuracy of two categories of TCP throughput prediction techniques. We then analyze the factors that affect the accuracy of each. 2. Congestion control and message loss in Gnutella peer-to-peer networks: We evaluate the congestion control mechanisms and message loss behavior in a real-world overlay network, the Gnutella system. The challenges for congestion control in such a network are analyzed, as are the design tradeoffs of alternative mechanisms. In order to study systems such as the above with details of the network, we build a scalable, extensible and portable packet-level simulator of peer-to-peer systems. The second part of the thesis, representing the second type of effort above, proposes two techniques to improve network simulation by exploiting the detailed knowledge of TCP: 1. Speed up network simulation by exploiting TCP steady-state predictability: We develop a technique that uses prediction to accurately summarize a series of packet events and, therefore, to save on processing cost while maintaining fidelity. Our technique integrates well with packet-level simulations and is more faithful in several respects than previous optimization techniques. 2. TCP workload generation under link load constraints: We develop an algorithm that generates traffic for a specific network configuration such that realistic and specific load conditions are obtained on user-specified links. At the same time, the algorithm minimizes the simulation memory requirement.