A replication system supports progressive update propagation by propagating a subset of the modified components and/or by propagating partial-fidelity versions of modified components.
In this section, we consider the implications of an implementation that supports progressive update propagation but does not support transcoding components on read or editing partial-fidelity components. In such an implementation, client replicas have by default full-fidelity versions of the components they replicate. A client replica has a partial-fidelity version for a component only when the component is being updated by some other client and the updates are being progressively propagated. In other words, the decision to propagate partial-fidelity data is made by the replica that is writing the component and not by the reader, as was the case in the previous section. Moreover, independently of whether we implement a pessimistic or optimistic approach to replication, once a partial-fidelity version has been propagated to the primary replica, it can only be replaced with another version created by the same writer (i.e., a higher-fidelity version or a more recent version). Replacing a partial-fidelity version with a version created by a different writer would require editing partial-fidelity components, which is not allowed by the implementation discussed in this section. In Section 2.3 we describe an implementation that allows a writer to replace a partial-fidelity version created by another writer.