Adaptive distributed XML views

Technical reportWe consider the problem of efficiently managing large corpora of XML documents in a distributed, decentralized setting involving hundreds of machines (or peers). We assume distributed users are willing to share their data. Thus, on any network peer, users may establish long-running subscriptions, by means of declarative queries over the entire current and future data. Thus, XML data is “liquid”: it comes from many sources and flows towards its consumers. Moreover, data accumulating at a peer as a result of a subscription is seen as a materialized view over the whole network, and can be used to answer other queries. We have developed ViP2P, a fully functional system implementing these ideas. We describe the architecture and the algorithms behind it, as well as experiments on hundreds of machines distributed in a LAN, in [7]. In this report, we consider the issue of adapting the materialized views established by a given ViP2P peer, to the needs of the peer itself, and of the other peers in the network. We outline the goals and techniques of adaptation, and discuss practical algorithm for evolving the set of views of a peer to better suit the needs of the peer, and of the whole network