Deriving CGM based-parallel algorithms for the optimal binary search-tree problem

Abstract

This paper presents a methodology to derive CGM (Coarse Grain Multicomputer) parallel algorithms for the cost of the Optimal Binary Search Tree Problem (OBST Problem). Depending on the parameter we want to optimize, we derive an algorithm accordingly. Therefore a load balancing, an efficient and a minimum communication rounds algorithms are obtained. Our CGM algorithms use p processors, each with O(n2/p) local memory. The best one in communication requires O(p1/2) communication rounds and O(n2/p) computations on each processor. Another one with the best time efficiency needs only O(n2/p) time steps. All local computations on processors are based on the Knuth sequential algorithm for the OBST problem.