Bulk-Loading The Nd-Tree In Non-Ordered Discrete Data Spaces

Applications demanding multidimensional index structures for performing efficient similarity queries often involve a large amount of data. The conventional tuple-loading approach to building such an index structure for a large data set is inefficient. To overcome the problem, a number of algorithms to bulk-load the index structures, like the R-tree, from scratch for large data sets in continuous data spaces have been proposed. However, many of them cannot be directly applied to a non-ordered discrete data space (NDDS) where data values on each dimension are discrete and have no natural ordering. No bulk-loading algorithm has been developed specifically for an index structure, such as the ND-tree. in an NDDS. In this paper, we present a bulk-loading algorithm, called the NDTBL, for the ND-tree in NDDSs. It adopts a special in-memory structure to efficiently construct the target ND-tree. It utilizes and extends some operations in the original ND-tree tuple-loading algorithm to exploit the properties of an NDDS in choosing and splitting data sets/nodes during the bulk-loading process. It also employs some strategies such as multi-way splitting and memory buffering to enhance efficiency. Our experimental studies show that the presented algorithm is quite promising in bulk-loading the ND-tree for large data sets in NDDSs.