Techniques de localisation et de résumé des données dans les systèmes P2P. (Data Localization and Summarization Techniques in P2P Systems)

The goal of this thesis is to contribute to the development of data localization and summarization techniques in P2P environments. At the application layer, we focus on exploiting the semantics that can be captured from the shared data. These semantics can improve the search efficiency, and allow for more query facilities. To this end, we introduce a novel data indexing technique into P2P systems that relies on linguistic summarization. Our summaries are synthetic, multidimensional views that support locating relevant data based on their content. More interestingly, they provide intelligible data representations which may return approximate answers for user queries. At the P2P network layer, we focus on exploiting the characteristics of the overlay topology, namely its clustering features, in order to reduce the traffic overhead generated by flooding-based mechanisms. This allows to improve the performance of P2P systems, irrespective of the employment of techniques relying on data semantics at the application layer. To this end, we define a cluster-based search technique which is implemented over a connectivity-based clustering protocol. A connectivity-based clustering protocol aims to discover the natural organization of nodes, based on their connectivity. Thus, it delimits the boundaries of non-overlapping subgraphs (i.e. clusters) which are loosely connected, and in which nodes are highly connected. In this thesis, we first survey P2P data sharing systems. We focus on the evolution from simple file-sharing systems with limited functionalities, to Peer Data Management Systems (PDMSs) that support advanced applications with more sophisticated data management techniques. Second, we propose a solution for managing linguistic summaries in P2P systems. We define an appropriate summary model and efficient techniques for summary creation and maintenance. We also discuss query processing in the context of summaries. Third, we propose a cluster-based search technique on top of existing connectivity-based clustering protocols. We focus on reducing redundant query messages which unnecessarily overload the system. We validated our solutions through simulation and the results show good performance.