Abstract
Spatial join is an important yet costly operation in spatial databases. In order to speed up the execution of a spatial join, the input tables are often indexed based on their spatial attributes. The quadtree index structure is a well-known index for organizing spatial database objects. It has been implemented in several database management systems, e.g., in Oracle Spatial and in PostgreSQL (via SP-GiST). Queries typically involve multiple pipelined spatial join operators that fit together in a query evaluation plan. In order to extend the applicability of these spatial joins, they are optimized so that upon receiving sorted input, they produce sorted output for the spatial join operators in the upperlevels of the query evaluation pipeline. This paper investigates the use of quadtree-based spatial join algorithms and how they can be adapted to answer queries that involve multiple pipelined spatial joins in a query evaluation plan. The paper investigates several adaptations to pipelined spatial join algorithms and their performance for the cases when both input tables are indexed, when only one of the tables is indexed while the second table is sorted, and when both tables are sorted but are not indexed.
