Replication for Mobile Computers
project developed several mobile data replication services.
The most mature of these was Rumor.
Rumor is a file replication system that can be used to keep multiple copies of files consistent. Rumor is especially useful for keeping replicated files on mobile computers. Maintaining multiple replicas of each file allows for higher availability. In particular, allowing mobile computers to store file replicas ensures that users will be able to access important data while disconnected, while still maintaining a single environment shared with workstation machines when connected. To further improve availability, Rumor uses an optimistic replication strategy. Optimistic replication permits any accessible replica to be updated, thus allowing disconnected portable computers to not only read, but also write, their local file replicas.
Rumor is a portable optimistic replicated file service built as a wrapper around an existing file system. A Linux version of Rumor is up and running, with a beta release available on the web. Rumor portability is achieved by dividing the system's functionality into a larger system independent replication service and a smaller system dependent service.
Rumor uses important algorithms and data structures originally built for Ficus. Ficus is a successful in-kernel replication service based on SunOS 4.1.1.
Since Rumor is a user-level system, it can be installed and maintained by normal users. No root privileges are required.
Rumor's optimistic replication strategy will permit concurrent updates which can lead to inconsistent replicas of a single file. Rumor detects such inconsistencies and applies automated mechanisms to resolve them.
We are currently working on other replication solutions to mobile computing problems, including a file hoarding service for portable computers (Seer) and a file replication service capable of supporting hundreds of writable replicas.
|The primary Rumor paper was
at the ER'98 Workshop on Mobile Data Access.
A position paper
on Rumor was presented in the SIGCOMM Workshop on Middleware.