Multi-Dimensional Indexes in DBMSs

Abstract

Multi-dimensional data is present across multimedia, data mining and other data-driven applications. The R-Tree is a popular index structure that DBMSs are implementing as core for efficient retrieval of such data. The gap between the best and worst-case performance is very wide in an R-tree. Thus, building quality R-trees quickly is desirable. Variations differ in how node overflow are approached during the building process. This article studies the R-Tree technique that the open-source PostgreSQL DBMS uses. Focus is on a specific parameter controlling node overflows as an optimisation target, and improved configurations are proposed. This parameter is hard-wired into the DBMS, and therefore, an implementation is presented to allow this parameter to become accessible through an SQL construct. The access method designer can resort to configuring this parameter when trying to meet specific storage or time-related performance targets. With this study, the reader can gain an insight into the effects of changing the parameter by considering the spatial indexes on well-known workloads.