Resequencing and Clustering to Improve the Performance of Spatial Joins

The lter-and-reene strategy is well-established as the basis for spatial join algorithms. In contrast to the lter step, the reenement step has received little attention, despite contributing signiicantly to the total cost of a join evaluation. Sorting candidate tuples as produced by the lter step has recently been shown to reduce the I/O cost of reenement. Our paper reports investigations of spatial join algorithms, with particular emphasis on interactions between choices of algorithms for the lter, sequencing and reenement steps and on the eeects of clustered and unclustered organization of full spatial descriptions of objects. We extend existing techniques by caching spatial descriptions and a new sorting strategy, called zigzag. Our experiments connrm that the choice of the sequencing strategy used is very important and that clustering has a signiicant innuence on join performance. Clustering by z-ordered keys is shown to be superior to ordering by minimum bounding rectangles.