Bayou

  The Bayou system [15] offers one solution for supporting the needs of DCAs. The goal of Bayou is to support distributed shared data applications with client-side updates in an environment with disconnected operation. Updates to shared state are propagated to clients using a variant of Golding's time-stamped anti-entropy (TSAE) protocol [5] for ordered group communication. This protocol enforces the policy of eventual consistency: in the absence of new updates, all caches converge to the same consistent state.

Our work has similar goals to Bayou. Both provide eventual consistency of cached data for distributed collaborative applications. However, Bayou's implementation of eventual consistency uses merge and rollback procedures to automatically handle conflicting updates to shared data. Keeping in mind that our work is designed for graphical applications, we have no notion of a conflict, therefore no need for merge or rollback procedures.

Another difference between Bayou and Relay is server topology. Bayou uses a peer-to-peer structure and TSAE to propagate updates. Peer-to-peer communication can improve scalability, but it is difficult to maintain when the set of peers is likely to change. In a peer-to-peer system, clients pair up to exchange updates. If the set of clients changes, a stranded client must be aware of other peers to continue with the TSAE protocol. DCAs with client/server communication can avoid this overhead and state. Scalability in a client/server system can be improved through a hierarchy of servers (Chapter 3).