Fraîcheur et validité de données répliquées dans des environnements transactionnels

We propose a framework for managing the quality of data replicated optimistically on a database cluster. It is based on a model of quality of data. Qualitalively, we make the difference between “freshness” and “validity” of data. Quantitatively, data quality is expressed through divergence measures between the data read and the same data with perfect quality. Users specify a minimum level of quality and the system uses it to elaborate the fastest satisfying execution plan. Our approach is a middleware that preserves application and database autonomy. We provide an experimental validation for mono-master replication configurations and freshness relaxing based on our prototype Refresco. The results show that freshness control can help increase query throughput significantly. They also show significant improvement when freshness requirements are specified at the relation level rather than at the database level. MOTS-CLÉS : qualité des données, réplication, cluster, fraîcheur, validité.