Metaheuristic based peer rewiring for semantic overlay networks. (Métaheuristique pour la configuration dynamique de réseaux pair-à-pair dans le context des réseaux logiques sémantiques)

A Peer-to-Peer (P2P) platform is considered for collaborative Information Retrieval (IR). Each peer hosts a collection of text documents with subjects related to its owner's interests. Without a global indexing mechanism, peers locally index their documents, and provide the service to answer queries. A decentralized protocol is designed, enabling the peers to collaboratively forward queries from the initiator to the peers with relevant documents. Semantic Overlay Network (SON) is one of the state-of-the-art solutions, where peers with semantically similar resources are clustered. IR can then be e ciently performed by forwarding queries to the relevant peer clusters in an informed way. SONs are built and maintained mainly via peer rewiring. Speci cally, each peer periodically sends walkers to its neighborhood. The walkers walk along peer connections, aiming at discovering more similar peers to replace less similar neighbors of its initiator. The P2P network hence gradually evolves from a random overlay network to a SON. Random and greedy walk can be applied individually or integrated in peer rewiring as a constant strategy during the progress of network evolution. However, the evolution of the network topology may a ect their performance. For example, when peers are randomly connected with each other, random walk performs better than greedy walk for exploring similar peers. But as peer clusters gradually emerge in the network, a walker can explore more similar peers by following a greedy strategy. This thesis proposes an evolving walking strategy based on Simulated Annealing (SA), which evolves from a random walk to a greedy walk along the progress of network evolution. According to the simulation results, SA-based strategy outperforms current approaches, both in the e ciency to build a SON and the e ectiveness of the subsequent IR. This thesis contains several advancements with respect to the state-of-the-art in this eld. First of all, we identify a generic peer rewiring pattern and formalize it as a three-step procedure. Our technique provides a consistent framework for peer rewiring, while allowing enough exibility for the users/designers to specify its properties. Secondly, we formalize SON construction as a combinatorial optimization problem, with peer rewiring as its decentralized local search solution. Based on this model, we propose a novel SA-based approach to peer rewiring. Our approach is validated via an extensive experimental study on the e ect of network rewiring on (i) SON building and (ii) IR in SONs. Cette thèse est accessible à l'adresse : http://theses.insa-lyon.fr/publication/2014ISAL0036/these.pdf © [Y. Yang], [2014], INSA de Lyon, tous droits réservés