Thèse soutenue

L'analyse et l'optimisation des systèmes de stockage de données dans les réseaux pair-à-pair

FR  |  
EN
Auteur / Autrice : Abdulhalim Dandoush
Direction : Philippe NainSara Alouf
Type : Thèse de doctorat
Discipline(s) : Informatique
Date : Soutenance en 2010
Etablissement(s) : Nice
Ecole(s) doctorale(s) : École doctorale Sciences et technologies de l'information et de la communication (Sophia Antipolis, Alpes-Maritimes)

Résumé

FR  |  
EN

This thesis characterizes the performance of peer-to-peer storage systems in terms of the delivered data lifetime and data availability. Two schemes for recovering lost data are modeled and analyzed: the first is centralized and relies on a server that recovers multiple losses at once, whereas the second is distributed and recovers one loss at a time. For each scheme, we propose several Markovian models that equally apply to many distributed environments as shown through numerical computations. These allow to assess the impact of each system parameter on the performance. In particular, we provide guidelines on how to tune the system parameters in order to provide desired lifetime and/or availability of data. The key assumptions made in the models are validated through intensive packet-level simulations or real traces collected from different distributed environments. In fact, we propose a realistic simulation model implemented on the Network Simulator (NS-2) for both download and recovery processes. Although this simulator can accurately predict the behaviour of the latter processes while considering the impact of several constraints such as the heterogeneity of peers and the the underlying network topologies, this simulator requires however relatively long time. To overcome this scalability limitation, we propose and analyze an algorithm. The algorithm is efficient in time and quite simple and uses the concept of ``Progressive-Filling'' (or max-min fairness). The validation of this algorithm consists in characterizing the distribution of the response time of parallel downloads in a distributed storage system, through simulations.