Résumé Motivation

Heterogeneity, at service and technology level, is one of the main characteristics of todays networks. The ever increasing offer of new services has changed not only the amount of traffic in the network but also its structure. As a result, traditional techniques for traffic prediction and classification are becoming obsolete. In the first part of the thesis, we address these two problems by means of advanced statistical tools. We analyze the problem of online prediction of the load on a link based only on past measurements and without assuming any particular model. Concerning traffic classification, and motivated by the widespread use of P2P systems, we focus on the identification of P2P applications, considering more precisely the case of P2P television (P2P-TV). For both cases, our framework makes use of Support Vector Machines (SVM). The algorithms we propose provide very accurate results, they are robust and their computational cost is extremely low. These properties make our solutions specially adapted to an online application. As traffic is increasingly heterogeneous, so are the access technologies. In particular, wireless is destined to be the access technology of choice. In this context, self-organized systems such as Mobile Ah-doc Networks (MANETs), are of particular importance. The design of an efficient and fair multiple access (MAC) mechanism is then crucial. In the second part of the thesis, we address two different problems related to MAC mechanisms in MANETs (in particular, we concentrate on CSMA since it is the most widely deployed mechanism). Firstly, given the absence of a mathematical model of consensus, an analysis of the existing models, with special emphasis on their correlation with the real protocol, is presented. Some weakness are identified and possible solutions are proposed. The use of stochastic geometry tools allows us to obtain analytical results where other techniques cannot. Secondly, we address the problem of lack of QoS in CSMA and we propose two different mechanisms that guarantee a minimum rate for each accepted transmission. Both take the interference level into account to decide on the set of connections which can access the shared channel at any given time. The main difference between them is the possibility or not of adjusting the transmission power of the nodes. The main aim of our study is to identify which of the proposed mechanisms outperforms CSMA best depending on the scenario. 5 pa st el -0 00 05 85 3, v er si on 1 8 M ar 2 01 0