Efficient parallel permutation-based range-join algorithms on mesh-connected computers

Abstract

This paper proposes three efficient parallel algorithms for computing the range-join of two relations on two-dimensional n x m mesh-connected computers, where n and m are the numbers of the rows and columns respectively. After sorting all subsets of both relations, all proposed algorithms permute all sorted subsets of one relation to each processor in the computers, where they are joined with the subset of the other relation at that processor by using a sequential sort-merge rangejoin algorithm. The Min-Storage-Shifting and Min-Movement-Shifting algorithms permute the data on a mesh alternatively in the row and column directions, and Hamiltonian-cycle algorithm permutes the data along a Hamiltonian cycle of the mesh. The analysis shows that the Hamiltonian-cycle algorithm requires fewer local join operations but more data movements than other two algorithms and that the Min-Movement-Shifting algorithm requires fewer local join operations and data movements but more storage than the Min-Storage-Shifting algorithm.