Fast Parallel Algorithms for Processing of Joins

Dennis E. Shasha, Paul G. Spirakis

Abstract: We present and analyze here some innovative techniques for processing a join (or a semi-join) in a parallel computing environment. Our algorithms employ perfect hashing and, in some cases, copying of data in a group of processors, or filtering the data as they move through the network. By using the combinatorial properties of hashing we are able to prove almost optimal speedup, with high probability, when some uniformity assumptions hold for the data. Even in the absense of these assumptions our techniques achieve sub-optimal speedup and can be used as practical heuristics.