Integrating spatial data management and object store technology

Abstract

Efficient access to spatial data is essential for many engineering, town planning and mining applications. Conventional database systems are not capable of performing such searches effectively over large data sets. Relational databases typically have to be used in conjunction with separate GIS (Geographical Information System) software, which is highly unsuitable from both system performance and programmer productivity viewpoints. There has recently been considerable research into building persistent object systems (POS) that can overcome the limitations of traditional databases. A POS allows data to be manipulated independently of its longevity; that is, program objects of any type can be made to persist on disk, and program statements have the same form whether they operate over transient or persistent data. A POS frees programmers from managing transfers between memory and disk, and from coding type conversions as data migrates between stores. It also reduces runtime costs because schema validation is not required on transaction execution due to the consistent binding between code and data. This paper describes a selection of spatial indexing mechanisms and their adaptation to run on persistent object stores. We studied two complementary techniques implemented on the Napier88 persistent store - namely grid files and R*-trees - and a generic spatial indexing library based on the GiST ( Generalized Search Tree) system for Persistent Java. Having measured performance over a variety of data sets and conditions, we conclude that persistent object systems are highly suited to spatial data management, and suggest guidelines for choosing a particular POS and spatial indexing method based on application-specific characteristics.