Differential Evolution Algorithm And The Research On Grid Resource Scheduling

Grid has become a hot Internet technology from the mid-90s till now. Its ultimate goal is to distribute hundreds of millions of resources together to eliminate information silos and realize the information sharing globally. However, the grid environment resources are geographically widely distributed, so resource management has become very difficult. One of the key lies in grid resource scheduling. Reasonable use of the algorithm to the resource scheduling among, not only can make full use of resources, but also between the allocation of resource and task ordered to greatly enhance the efficiency of task execution.Based on the problem of grid resource scheduling differential evolution algorithm is proposed, to better address the resource and task allocation. The concept, characteristics and architecture of grid is given in detail in this paper, on which the various characteristics of grid resources is described, mainly in scheduling principles, characteristics and scheduling process. What's more, the paper detail described the differential evolution algorithm, including its concepts, principles and implementation process, and give polynomial differential evolution algorithm applied specific experiment. Besides, analyzed the advantages differential evolution algorithm compared to other evolutionary algorithms.This paper proposed the improvement of algorithm on grid resource scheduling model. Improvements can be divided into three areas, first is for the improvement of the initial population. Use the selection of the elite population, which can ensure high performance of individual genetic. The second improvement is for the mutation in the scaling factor. The traditional scaling factor is a fixed value within a certain range. If the scale factors initially set too small, the population evolution is easy to fall into local features, too big, in later stage of evolution the convergence speed will too slow. The proposal of adaptive scaling factor can make a larger value in the early evolution for the strong global search ability, and a smaller value in the later stage of evolution to accelerate the convergence. The third point is a combination of improved particle swarm optimization. Mainly used the way of individual and global excellent value in the mutation in which the individuals can maintain the population excellent genetics as much as possible and the convergence rate was enhanced greatly.The major codes of three improvements are described with analysis of their experimental results, and then they were combined together. The experiment shows that algorithm was significantly improved compared with before, through which it can be seen that the algorithm to solve the problem of grid resource scheduling is feasible, and better than the performance of the original algorithm.