Semantics based transaction management techniques for replicated data

Data is often replicated in distributed database applications to improve availability and response time. Conventional multi-copy algorithms deliver fast response times and high availability for read-only transactions while sacrificing these goals for updates. In this paper, we propose a multi-copy algorithm that works well in both retrieval and update environments by exploiting special application semantics. By subdividing transactions into various categories, and utilizing a commutativity property, we demonstrate cheaper techniques and show that they guarantee correctness. A performance comparison between our techniques and conventional ones quantifies the extent of the savings.