Lifetime-based dynamic data replication in P2P systems

Using large-scale infrastructures such as grid or p2p systems for enhancing database replication performances is a tremendous solution because it avoids bottlenecks and single point of failure. One of the widely used replication policies is master-slave configuration which eases consistency control since all update operations are handled on a particular node, called master, whereas read operations are routed on remaining nodes. This approach improves read operations performances (i.e. low latency and high throughput) and therefore is well suited for read-intensive applications such as web 2.0 applications. However, replicating data on a large-scale system is very challenging due to nodes dynamicity that can compromise consistency and availability. To face this problem, we propose a dynamic data replication model, which manages efficiently nodes in such a way that consistency is always kept and data is available for read or write operations. To this end, we rely on the estimated lifetime of nodes in order to manage master node failures in a preventive fashion. Simulation results obtained show the feasibility of our approach and its positive impact on the global performances of the system.