On Optimal Node Splitting for R-trees

The problem of finding an optimal bipartition of a rectangle set has a direct impact on query performance of dynamic R-trees. During update operations, overflowed nodes need to be split (bipartitioned) with the goal of minimizing resultant expected query time. The previous algorithm for optimal node splitting requires exponential time. One contribution of this paper is a polynomial time algorithm for finding optimal bipartitions for any objective function whose value depends exclusively on the bounding hyper-rectangles of the ensuing partitions. The algorithm runs in O(nd) time where d > 1 is the number of dimensions of the input. Experimental studies indicate that the use of optimal splits alone results in improvements of query performance of only between 5% and 15% when compared to other heuristics. Thus, a second contribution is to demonstrate the near optimality of previous split heuristics, a fact that suggests that research should focus on global rather than local optimizatio...

• Type: Conference
• Year: 1998
• Where: VLDB
• Authors: Yván J. García, Mario A. Lopez, Scott T. Leutenegger
• Keywords: Database, Optimal Bipartitions, Optimal Splits, Query Performance, VLDB 1998