A Self-Adaptive Genetic Algorithm for Tasks Scheduling in Multiprocessor System

Abstract

Task scheduling is one of the crucial issues to achieve high performance for parallel multiprocessor systems. With the extensive studies of the task scheduling problem, many new methods, especial genetic algorithms, have been introduced into this field. In this paper, we develop a novel genetic algorithm, namely the self-adaptive genetic algorithm (SAGA). SAGA is different from the previously proposed genetic algorithms in a number of ways. Unlike the other genetic algorithms, SAGA makes some key parameters changeable with variable policy over the evolution. SAGA also efficiently generates the initial population, which may contain any possible feasible solutions. Simulation results show that SAGA outperforms the previously proposed algorithms in terms of the solution quality.