Direção de pesquisa : Marcelo DIAS DE AMORIM
The market penetration of mobile devices, their hardware capacities, as well as the multiplicity of available applications have experienced an impressive growth in the latest years. As a consequence, smartphones, tablets, and laptops have become both producers and consumers of user-generated contents. They also motivate novel communication paradigms such as the possibility to establish, in an opportunistic fashion, direct device-to-device links whenever two mobile nodes enter within the wireless range of each other.
Value-adds of opportunistic communications range from extended coverage where there is no network infrastructure available to the realization of offloading strategies to help operators relieve the load in their infrastructures. In this thesis, we consider the case of opportunistic dissemination of multiple large contents from an experimental point of view. This implies revisiting, among others, the common assumption that contacts have enough capacity to transfer any amount of data.
In the first part of this thesis, we start from an Android implementation of EPICS, a network protocol especially designed for exchanging large contents in opportunistic networks, on off-the-shelf devices. We conduct an experimental campaign evaluating its performance in many scenarios. Although the results revealed good performance for a few contents, the system shows severe limitations when scalability comes at play. Unfortunately, application-level logs stored during the experimentation are not enough to understand unexpected results. We propose then to rely on passive traffic monitoring and wireless traces analysis to find out limitations and uncovered improving possibilities. This methodology suggests us to better exploit the dynamics of the network topology through DAD, a new content dissemination protocol that adaptively sends bursts of data instead of the per-fragment transmission strategy of EPICS. We compare both protocols experimentally and using synthetic contact traces and show significant gains of the proposed approach.
Passive monitoring is an essential part of our work and we decided to investigate further some issues that remained open when we performed our experiments. The second part of this thesis deals with how to tackle the scalability problem of legacy WLAN monitoring systems in order to conduct a wide area experimental measurement. We propose two original approaches. With the first one, based on trace similarity and community detection algorithms, we are able to identify how many monitor we need in a target area and where to place them. On the other hand, given a fleet of monitors, the same method can be used to stretch the area under observation. The second approach in based on collaborative measurements. In this case we face the risk of biased measures due attacks of malicious users generating adulterated traces. We then propose a method to detect such malicious behaviors by using graph-based analysis of collected traces.
Defesas : 28/05/2014 - 14h - Site Jussieu 25-26/105
Andrzej Duda,Professeur Grenoble INP-Ensimag [Rapporteur]
Thierry Delot, Professeur Université de Valenciennes [Rapporteur]
Olivier Marcé, Ingénieur de Recherche Alcatel-Lucent Bell Labs
Luigi Iannone, Maître de Conférence Télécom ParisTech
Giovanni Pau, Professeur UPMC Sorbonne Universités
Marcelo Dias de Amorim, Directeur de Recherche CNRS