Deferred maintenance of replicated objects in single-site databases

Replication of database objects is suggested as a means of enhancing performance in cases where conflicting preferences for clustering occur, such as in the storage of many-to-many relationships. A novel feature is lazy propagation of updates: the primary objects are kept up-to-date, but an obsolete replica is not updated until it is next retrieved. The core of the method is a main-memory hash table, containing timestamps of primary objects. Replication causes almost no overhead in regard to updates, concurrency or recovery. Yet, the retrieval speed can be increased considerably. Replication should be applied only for critical, read-intensive object types, to gain actual advantage from redundancy, compensating the storage penalty. The author´s scheme could be used for implementing materialized views in relational databases. However, the method is described in terms of pointer-based architectures