Parallelization by the divide-and-conquer method
Qi Yang, C. Yu · 2003
A precise characterization for problems to be parallelized by the divide-and-conquer method is obtained. A parallel algorithm for such problems, which is likely to cluster data such that a substantial amount of computation is carried out within sites and communication cost is manageable, is developed. The characterization turns out to be very simple, yet general enough to cover concrete problems in many disciplines, such as sorting, computing the transitive closure of a binary relation, and computing a minimal cover in decomposing relations into 3NF forms. A linear recursive program is given to describe problems to be parallelized.>