An E cient Data Replication Management ArchitectureExploiting

The management of replicated data in a distributed system is a diicult task. In order to ease the pain, we developed an architecture for data replication whose key concept is the separation of policy and mechanism. One of the major advantages of such an architecture is the fact, that { once the mechanism is implemented { new data replication strategies can be introduced into the network just by specifying the new policy. The resulting strategy, assumed that the general mechanism is already implemented, then is immediately executable. We brieey explain a exible framework for data replication strategies called General Structured Voting which is part of our architecture and is based on such a separation. Additionally, we demonstrate that the speciication of policies can be highly supported by using suited tools: a graphical editor allows the interactive design of new policies by a human expert whereas an automatic designer works independently once the properties of the desired policy are deened.