Adaptive Processing of Multi-Criteria Decision Support Queries.

Business intelligence applications ranging from stock market tickers to strategic supply chain adaptation systems require the efficient support of multi-criteria decision support (MCDS) queries. Skyline queries are a popular class of MCDS queries that have received a lot of attention recently. However, a vast majority of skyline algorithms focus entirely on the input being a single data set. In this work, we instead focus on supporting the more powerful SkyMapJoin queries. Our Adaptive-SKIN framework conducts processing at two levels of abstraction thereby effectively minimizing the two primary costs, namely the cost of generating join results and the cost of dominance comparisons to compute the final skyline of the join results. Our proposed approach hinges on two key principles. First, in the input space - Adaptive-SKIN determines the abstraction levels dynamically at run time instead of assigning a static one at compile time. This is achieved by adaptively partitioning the input data driven by the feedback of the results already generated. Second, Adaptive-SKIN incrementally build the output space, containing the final skyline, without generating a single join result. Our approach selectively drills into regions in the output space that show promise in generating result tuples as well as avoiding the generation of intermediate results that do not contribute to the query result. In this effort, we propose a cost-vs.-benefit driven strategy for abstraction selection. Our experimental evaluation demonstrates the superiority of the Adaptive-SKIN over state-of-the-art techniques over benchmark data.