Performance Evaluation of Parallel Divide-and-Conquer Algorithms

this paper we evaluate the performance of several parallel divide--and--conquer algorithms which are created in a uniform manner by a software system developed for automatically transforming a sequential divide--and--conquer algorithm into parallel code, instead of individually designing a parallel algorithm by hand. The system partitions a sequential algorithm programmed in C into independent tasks, maps these to a MEIKO transputer system and executes them in parallel. From the speedups obtained it is evident that not all algorithms considered are worth to be parallelized, and we will present additional measurements to explain why. 2 General Approac