Abstract
The optimization of the response time in a parallel pipelined join of n relations is formally defined and analyzed. Multiple data-flow streams in the pipeline are captured in a parallel processing tree (PPT) that models both the parallel and the sequential aspects of the pipeline. An optimal PPT achieves minimal first-page response time by balancing the data flow in the pipeline. It is shown that the optimal PPT problem is NP-complete and that only for a very limited set of queries can one obtain optimal PPT. The authors then provide heuristic algorithms for tree and nontree queries that construct a suboptimal PPT. The dynamic feature of the PPT is developed. Experiments have been conducted to verify the performance of the PPT obtained by the heuristic algorithm.<>