Boolean Bounding Predicates for Spatial Access Methods
DEXA '02 Proceedings of the 13th International Conference on Database and Expert Systems Applications
| Hi-index | 0.00 |
Tree-based indexes are integral to efficient querying in multimedia and GIS applications. We propose algorithms to reduce the number of I/Os performed in these common data structures. Multidimensional indexes frequently use shapes to describe the data stored in the subtree below a given rectangle. For example, the popular R*-tree uses minimum bounding rectangles. Our approach creates boolean combinations of rectangles to replace the original rectangles if there is free space on the node to do so. We implemented and tested several algorithms for deciding which rectangles in an index node to replace with multiple rectangles, and implemented and tested several methods for deciding which nodes in the index to improve when available CPU time is limited. We ran experiments over a variety of real and synthetic data sets, examining the performance benefit of the various index node tuning algorithms. Experiments show that up to fifty percent of the unnecessary I/Os caused by imprecise data descriptors can be eliminated using our algorithms.