Multi-Objective Query Processing for Data Aggregation

Most fielded data integration systems focus on data aggregation applications, where individual data sources all export fragments of a single relation. Given a query, the primary query processing objective in these systems is that of selecting the appropriate subset of sources so as to optimize various user objectives regarding the completeness and quality of the answers and the response time. In this paper we consider three specific objectives: coverage, density and latency. To handle the often conflicting nature of these objectives, we introduce a joint optimization model for source selection that supports a spectrum of trade-offs between them. We also introduce our techniques in defining query classes for different types of statistics and learning statistics with respect to corresponding query classes from the autonomous data sources. We present comprehensive evaluation to demonstrate the effectiveness of our multi-objective query processing model for data aggregation scenarios.