USE OF THE GRASP METHODOLOGY FOR TASKS ALLOCATION IN MIMD PARALLEL COMPUTERS

摘要

In this article we extend the methodology proposed in to allocate tasks in a parallel computer by means of the GRASP methodology. Algorithms are represented by tasks graphs with precedence relationship. Two characteristics of the graphs are exploited: the loops of the tasks and the partial order. The allocation of the tasks is analyzed as an optimization problem where our objective was to minimize the communications between the tasks and also reduce the idle time. This methodology is applied to the Jamenson algorithm that solves the Euler equations, obtaining an efficient parallel implementation.