Replication combined with explicit management of data integrity constraints can be used to enhance availability of object-oriented, data-centric distributed systems when node and link failures occur. Our approach enhances availability by temporarily relaxing non-critical data integrity constraints during degraded situations. This requires new kinds of optimistic replication protocols that support the configuration of this trade-off. The contribution of this paper is a replication model called Availability/Consistency Balancing Replication Model that allows replicas to diverge in degraded situations if data integrity can be temporarily relaxed and re-establishes both replica consistency and data integrity during repair time. The Primary-per-PartitionProtocol and Adaptive Voting are two concrete protocols following our model. The feasibility of our approach has been shown by several prototype implementations.
Johannes Osrael, Lorenz Froihofer, Karl M. Gö