Plug&join: An easy-to-use generic algorithm for efficiently processing equi and non-equi joins

Abstract

This paper presents Plug&Join, a new generic algorithm for efficiently processing a broad class of different join types in extensible database systems. Depending on the join predicate Plug&Join is called with a suitable type of index structure as a parameter. If the inner relation fits in memory, the algorithm builds a memory resident index of the desired type on the inner relation and probes all tuples of the outer relation against the index. Otherwise, a memory resident index is created by sampling the inner relation. The index is then used as a partitioning function for both relations. In order to demonstrate the flexibility of Plug&Join, we present how to implement equi joins, spatial joins and subset joins by using memory resident B+-trees, R-trees and S-trees, respectively. Moreover, results obtained from different experiments for the spatial join show that Plug&Join is competitive to special- purpose methods like the Partition Based Spatial-Merge Join algorithm.