Contents   Index

Replication transparency

When introducing mobile computers into an otherwise stationary distributed system, it must be decided whether the existing applications should continue to function without any changes to them or whether new applications must be developed.

No existing applications (that I am aware of) use multi-level consistency, varying their consistency demands after the characteristics of current communication media. So existing applications must be rewritten (or taken special care of, see Section 5.8) if they are to use such a scheme. It could be decided to develop a whole new environment for mobile computing and write new applications which could make way for an optimal solution. But since the system I am going to develop is to be used in connection with an existing distributed system, then that is really not feasible. A solution that requires only small changes to existing applications is maybe the thing to decide upon, and that I will do.

Coda, Ficus, and MIo-NFS all have replication transparency, that is, the caching of files on the mobile computers, conflict detection, and conflict resolution is handled by system (as much as possible). Bayou on the other hand has absolutely no replication transparency; applications must be written with awareness of the fact that they are working on replicas and supply conflict detection and resolution.

A funny consequence of replication transparency is that when the system fails to resolve a conflict then the resolution of it is not left to the application--because it does not know about the replication--but to the users! So this is a case were transparency on application level leads to non-transparency on user level.

  Contents   Index