Comments on "Hash-based and index-based join algorithms for cube and ring connected multicomputers" by E.R. Omiecinski and E.T. Lin

The cost distributions of both the parallel hybrid-hash join and the parallel join-index join algorithms proposed in the above-named work (ibid., vol.1, p.329-43, Sept. 1989) are presented in more detail. The result shows that almost the entire relation may need to be retrieved from disk, though the join selectivity is low. A table of semi-join selectives and cube sizes is given to show the condition that the join-index method performs better than the hybrid-hash method, i.e., the really low selectivity for the join-index method. An error in one of the cost formulas is corrected, and a more efficient method on the final join in the join-index method is proposed.<>