Research And Implementation Of Grid Task Scheduling Based On Improved Evolutionary Algorithms

Gird system is currently most widely used distributed application system. It makes use of Internet to connect varieties of resources with geographical distribution into a virtual organism which is similar to a logical whole or a super computer, and the users can not only enjoy integrative information and application services, but also share resources and do collaborative work. The "resource island" phenomenon doesn’t exist in Grid and it accomplishes full sharing of information.The computing capability of Grid System is reflected by the running performance of grid tasks, in grid environment, task scheduling is one of the pivotal problems in the grid research field. However, due to dynamic, distribution and heterogeneous characters of grid environment, task scheduling is facing a tremendous challenge. On the basis of analyzing the process, targets and characteristic of task scheduling, this paper emphatically has done the following works:1. An improved coarse-grained parallel genetic algorithm is proposed on the basis of introducing implementation process, advantages and disadvantages of genetic algorithm. The algorithm designs a multi-point crossover operator based on schema order in the crossover phase and adopts a vectored mutation method based on task migration in mutation phase. What’s more, it uses elitist strategy to keep the population diversity. In addition, the theory of moving average in statistics is also introduced into the improved algorithm to predict the changing tendency of the population fitness after several generations in the evolution.2. An improved immune algorithm is proposed on the basis of introducing implementation process, advantages and disadvantages of immune evolutionary algorithm. The immune mechanism is introduced into the improved algorithm and the regulator k is adjusted to be a adaptive function. Simulation experiment shows that the improved algorithm can ameliorate the disadvantages of the algorithm improved before and keep the population diversity better. 3. The experimental results indicate that the algorithm proposed in this paper not only have quicker convergence rate and better optimization capability but also can obtain better scheduling results than the algorithm improved before.