Clone Join and Shadow Join: Two Parallel Algorithms for Executing Spatial Join Operations

With the growing popularity of spatial applications, there has been a signiicant increase in the use of database systems for storing and querying spatial data. Spatial data is now readily available from a variety of sources including government mapping agencies, commercial sources, satellite images, and simulation outputs. As this trend continues, applications continue to execute increasingly complex queries on large and larger volumes of spatial data. As can be expected, these complex spatial queries frequently involve joining two data sets based on some spatial relationship between objects in the two data sets. This operation is called a spatial join, and like its relational counterpart, is an expensive operation. Consequently, spatial database systems must employ eecient spatial join algorithms. In the past, many algorithms have been proposed for evaluating a spatial join operation on a single processor system. However, the use of parallelism for handling queries involving large volumes of spatial data has received little attention. In this paper, we explore the use of parallelism for storing and querying large volumes of spatial data. We rst propose and analyze some strategies for storing spatial data in a parallel database system. In this paper, we focus primarily on the static space partitioning approach that rst statically partitions the underlying space, and then maps these partitions to processors. We propose a number of spatial join algorithms based on these declustering strategies. Two algorithms are identiied as the primary algorithms in this design space. We develop analytical cost models for these two algorithms, and, using the analytical model we identify key parameters that innuence the performance of these join algorithms. Finally, using real data sets and an actual implementation, we test the performance of these algorithms. The experiments show that both algorithms can eeectively exploit parallelism.