For The Application Of Parallel Genetic Algorithms In Weight Programming Problem

With the rapid development of Internet technology on a global scale, the link traffic is almost doubling every year. In the near future, it is foreseeable that the data traffic boom will be coming due to the expansion of Internet technology for different purposes. These changes prompted many researchers to seek a more efficient method of network resources.OSPF protocol is one of the most widely used IGP protocol through a set of link weight values to control the routing and set the network traffic distribution. So the optimization of weight to the network link has a direct impact on the link load balancing and resource effective using. In recent years, there are various types of heuristic search algorithm to solve the problem of setting weights when the network planning, also in special circumstances, such as network constraints and the single link failure, those algorithms show good performance. However, with the continuous improvement of the practical problems of scale and complexity, coupled with the increasing complexity and expansion of the network topology, the time consumed by the general search algorithm in the search process will be doubling which is difficult to meet the growing needs by humans.This paper studies about the parallel genetic algorithm in the weight setting problem. We will reduce the time consumption of the general serial genetic algorithm through the parallelization of the algorithm to improve the performance of the algorithm’s time.In the second chapter we mainly study the general weights setting problem. Based on the existing framework of the genetic algorithm, there are two improvements:First, we combined the thought of the local optimization with the genetic algorithm to improve the performance of the algorithm, and changed an individual in the initial population in order to prevent the weight of the cross-border assignment. Secondly, we use the coarse-grained parallel genetic algorithm to improve the time performance. Then we analyzed the impact of the parallel parameters of the algorithm which increased after parallel. The third chapter is aimed at the weights setting problem over the QoS constraint. This chapter first introduces three Strategies to deal with the individual of the genetic algorithm which satisfy the constraint. Then we proposed a repair idea of the weights setting problem, and combined it with punishment strategies to improve the algorithm performance. Finally, we increase the times of repair in the coarse-grained parallel genetic algorithm to improve the performance of the algorithm. It solves the contradiction between repair frequency and algorithm execution time.In the fourth chapter, we mainly study the weight setting problem over the single link failure. Because of considering the link failure factors in the algorithm, it must calculate all the failure of each link of the network status. It will spend much time on the calculation of the fitness function. For this specific issue, we use the master-slave parallel model to improve the algorithm time performance. Finally, we propose the blended model which combines the coarse-grained parallel mode with the master-slave model to improve performance.