02/01/1993 09:00 AM
Computer Science
Distributed database systems for wide-area networks must scale to very large numbers of replicas in order to provide acceptable availability and response time. Weak-consistency replication protocols, such as the *timestamped anti-entropy* (TSAE) protocol we have developed, allow a database to scale to hundreds or thousands of replicas. The TSAE protocol allows updates to be processed by a single replica, then propagated from one replica to another in the background, causing replicas to temporarily diverge. The divergence is resolved in a short time and is resolved correctly even in the presence of temporary replica failure and network partition. We present a detailed analysis of the update propagation latency and the divergence between replicas caused by this protocol.
