The Case for Non-transparent Replication: Examples from Bayou.

Applications that rely on replicated data have different requirements for how their data is managed. For example, some applications may require that updates propagate amongst replicas with tight time constraints, whereas other applications may be able to tolerate longer propagation delays. Some applications only require replicas to interoperate with a few centralized replicas for data synchronization purposes, while other applications need communication between arbitrary replicas. Similarly, the type of update conflicts caused by data replication varies amongst applications, and the mechanisms to resolve them differ as well. The challenge faced by designers of replicated systems is providing the right interface to support cooperation between applications and their data managers. Application programmers do not want to be overburdened by having to deal with issues like propagating updates to replicas and ensuring eventual consistency, but at the same time they want the ability to set up appropriate replication schedules and to control how update conflicts are detected and resolved. The Bayou system was designed to mitigate this tension between overburdening and underempowering applications. This paper looks at two Bayou applications, a calendar manager and a mail reader, and illustrates ways in which they utilize Bayou’s features to manage their data in an application-specific manner.