| Grid is a set of emerging technologies built on Internet; it can integrate the resources all over the world to achieve the various resource sharing. So the basic goal of Grid is to realize share of heterogeneous resources. What the Grid task scheduling cares about is the output of the whole Grid system. But the resources in Grid are distributed geographically, heterogeneous inherently, autonomous, dynamic and so on which makes the task scheduling in Grid more complicated than in traditional environment, additionally task scheduling is a NP-complete problem in nature, So researching and presenting an algorithm which could schedule the task in Grid suitably and efficiently is so important.Based on the researching of Grid Computing and the situation of task scheduling, the paper chooses Adaptive Genetic Algorithms (AGA) to do the task scheduling of Grid. But AGA has a problem of premature convergence, in order to solve this we improves the AGA with small computing operations. In the process of population initialization a new method based on combining of the min-min algorithm and the mutation of Genetic Algorithm is addressed; in the evolution of the population a new criterion predicting the premature convergence is presented and the corresponding improved mutation is designed when premature convergence occurs; in the end a algorithm of neighborhood searching is used to seek the better solution. These operations are integrated in the original AGA. Then the paper conducts a deep research of GridSim running principle and extends the GridSim based on it. At last experiments validating the two methods are carried out in the extended GridSim. Comparing the experimentation results of the proposed algorithm with the original AGA, the proposed algorithm enhances global searching ability and quickens convergent speed. It shows great effectiveness in Grid task scheduling especially for the situation that the number of tasks is big. |
PDF Full Text Request