Conflict Detection in DeeDS

Abstract: In distributed database systems, immediate global consistency of replicated data can be achieved by distributed commit protocols that are typically unpredictable. If real-time characteristics are necessary, such unpredictability has to be avoided. In a distributed real-time database, optimistic replication can be used to avoid unpredictable delays by allowing transactions to commit locally. The update of other nodes is performed as soon as possible. If optimistic replication is used, conflicts may occur since data can be changed locally without synchronously informing other nodes. To detect these conflicts, this thesis introduces a conflict detection approach for DeeDS, a distributed, active real-time database that supports a dynamic node set. A comparison of existing conflict detection approaches is performed, and it is found that the dynamic version vector approach is the best fitting approach. The main reason is that it can handle a dynamic node set with a minimum of additional conflict detection data. To show the realization of the approach in DeeDS, dynamic version vectors have been implemented. Additionally, conflict management in DeeDS is redesigned to allow separation of conflict detection and conflict resolution. This makes the software architecture more flexible and is a first step towards application specific conflict resolution.