A new genetic-based algorithm for scheduling static tasks in homogeneous parallel systems

摘要

Task scheduling on multiprocessor systems is of a great interest since it is a computationally difficult problem due to having the possibility of failing in exploiting the true potential of the multiprocessor system as a result of inappropriate scheduling of available tasks. Static scheduling using genetic algorithms is a very popular approach as of being able to efficiently use the available machines and resources. In this paper, we propose an efficient algorithm, namely, sequence algorithm as a novel extension to a traditional existing generic algorithm to solve the task scheduling problem in multiprocessors systems through minimizing the tasks completion time and maximizing the throughput of the system. To achieve that, our proposed algorithm has its own way to initialize the chromosomes. Moreover, it uses a new systematic method for the crossover operator. In our work, we ignore the communication delay between tasks. To show the effectiveness of the proposed algorithm, we performed experimental simulations to see how far we need to go, for reproducing generations, to obtain acceptable efficiency. Actually, the results show that we can obtain a variance less than 10% (efficiency ≥ 90%) with a small number of iterations. By comparing the proposed algorithm with an existing genetic-based algorithm, it is found that the efficiency of the new algorithm to find a suboptimal schedule is increased for a fixed number of generations which in turn leads to decreasing the length of schedule or the completion time.