Scalable QSF-Trees: Retrieving Regional Objects in High-Dimensional Spaces

Many database applications require effective representation of regional objects in high-dimensional spaces. By applying an original query transformation, a recently proposed access method for regional data, called the simple QSF-tree (sQSF-tree), effectively attacks the limitations of traditional spatial access methods in spaces with many dimensions. Nevertheless, sQSF-trees are not immune to all problems associated with high data dimensionality. Based on the analysis of sQSF-trees, this paper presents a new variant of sQSF-trees, called the scalable QSF-tree (cQSF-tree), which relies on a heuristic optimization to reduce the number of false drops into pages that contain no object satisfying the query. By increasing the selectivity of search predicates, cQSF-trees improve the performance of multidimensional selections. Experimental evidence shows that cQSF-trees are more scalable than sQSF-trees to the growing data dimensionality. The performance improvements also increase with more skewed data distribution.