Réplication asynchrone des données dans trois contextes: entrepôts, grappes et systèmes pair-à-pair

摘要

In this HDR thesis, we present our contributions to improve data freshness and to manage strong consistency in single master and multi-master lazy replication configurations, respecting the autonomy of the database internals. This means that all the components necessary to support our protocols are implemented outside de DBMS. These contributions were motivated by distributed database system applications such as small scale Oneline Analysis Processing (OLAP) and small and large scale Oneline Transaction Processing (OLTP) in database cluster systems. Large-scale distributed collaborative applications are getting common as a result of rapid progress in distributed technologies (grid, P2P, and mobile computing). As an example of such applications, consider a second generation Wiki that works over a peer-to-peer (P2P) network and supports users on the elaboration and maintenance of shared documents in a collaborative and lazy manner. For simplicity, we consider three main classes: unstructured, structured, and super-peer networks. P2P systems allow decentralized data sharing by distributing data storage across all peers of a P2P network. Since these peers can join and leave the system at any time, the shared data may become unavailable. To cope with this problem, a solution is to replicate data over the P2P network. Several data replication solutions have been proposed in P2P systems to improve availability within the overlay network. Most of these solutions are based on lazy single master replication built for file systems. On the other hand, very few replication solutions have been proposed to handle the application requirements. Our research contributions for managing lazy multi-master data replication in P2P systems are related to specific collaborative applications. We adopt optimistic replication due to its flexibility which is necessary in dynamic and large scale environments. The first contribution in this subject is a P2P Topology Aware Semantic Reconciliation engine. Another important contribution is related to the improvement of performance on providing data availability in structured P2P systems in the presence of multiple replicas of a given object, given dynamicity and failures. In this context, we proposed a data replication service that uses the concept of currency over a distributed timestamp protocol.