Managing Change in Large-Scale Data Sharing Systems

The problem of sharing data across multiple sources has received considerable attention in recent years because of its relevance to enterprise data management, scientific data management, and information integration on the WWW. However, the management of updates in such systems has received very little attention. In a data sharing system, the set of sources and clients is not fixed, and therefore the sources publishing the updates do not necessarily know exactly who will consume them. Consequently, the system needs to support a variety of update propagation strategies. In this paper, our approach is based on identifying two kinds of objects of interest, which are treated as first-class citizens in the system: updategrams, which are descriptions of updates over base relations, and boosters, which complement updategrams to speed up the processing of join views. We derive a complete set of rules governing the production, combination, and reconciliation of updategrams and boosters over differing time intervals. Our rules cover both GAV and LAV style mediation, as well as views involving certain forms of aggregation. We show how to use our rules to produce efficient query execution plans by extending the System-R style query optimizer, and present experiments that evaluate several heuristics for pruning the search space of plans that use updategrams and views for query evaluation.