Real-Time Control of a Pipelined Multicomputer for the Relational Database Join Operation

We propose a database system which consists of a personal computer, disks and a pipeline of microcomputers. The rela tional join is a target. We find that there exist two different algorithms: one contributes the speed up of the pipeline cycle time, and the other reduces the number of the necessary disk accesses. The controller watches the state of the pipeline, and if the pipeline is critical, then it switches to the high speed algorithms. Like wise, the controller selects the most appropri ate strategy for the given situation. We have built an experimental parallel processing sys tem. The results show the feasibility of the proposed control algorithm.