OraGiST - How to make user-defined indexing become usable and useful

Abstract

In this article we present a concept for simplification of user-defined indexing for user-defined data types in object-relational database systems. The concept is based on a detailed analysis of user-defined indexing in ORDBS on one hand, and features of generalized search trees (GiST) as an extensible indexing framework on the other hand. It defines a minimal interface to be implemented in order to use GiST within ORDBS; this greatly simplifies the process of implementing user-defined indexes. The effectiveness of the approach is illustrated by performance experiments carried out on a prototypical implementation of our concept. For the experiments we have used new specialized spatial data types, that store spatial as well as thematic information within a single attribute. These data types facilitate advanced spatial analysis operators. The experiments show great performance improvements on these operators by using multidimensional user-defined index structures based on R-trees when compared to system-provided indexes.