Cost-based query-adaptive clustering for multidimensional objects with spatial extents. (Groupement d'Objets Multidimensionnels Etendus avec un Modèle de Coût Adaptatif aux Requêtes)

We propose a cost-based query-adaptive clustering solution for multidimen-sional objects with spatial extents to speed-up execution of spatial range queries (e.g.,intersection, containment). Our work was motivated by the emergence of many SDIapplications (Selective Dissemination of Information) bringing out new real challengesfor the multidimensional data indexing. Our clustering method aims to meet severalapplication requirements such as scalability (many objects with many dimensions andwith spatial extents), search performance (high rates of spatial range queries), updateperformance (frequent object insertions and deletions), and adaptability (to object andquery distributions and to system parameters). In this context, the existing indexing so-lutions (e.g., R-trees) do not efficiently cope with most of these requirements. Our objectclustering drops many properties of classical tree-based indexing structures (tree heightbalance, balanced splits, minimum object bounding) in favor of a cost-based clusteringstrategy. The cost model takes into account the performance characteristics of the exe-cution platform and relies on both data and query distributions to improve the averageperformance of spatial range queries. Our object clustering is based on grouping spatialobjects with similar intervals (positions and extents) in a reduced subset of dimensions,namely the most selective and discriminatory ones relative to the query distribution. Thepractical relevance of our clustering approach was demonstrated by a series of experi-ments involving large collections of multidimensional spatial objects and spatial rangequeries with uniform and skewed distributions.