Grid Resource Scheduling Based On Prediction Mechanism

Grid Technology is a hot technique developed in recent years; it makes full use of all kinds of geographically separated resources to form a super computer. As the development of Grid Technology, Grid resource scheduling has become more and more important. There are two aspects in the Grid Resource Scheduling Algorithms: one is to predict the resource state; the other is to schedule resources based on the result of the prediction. There are some disadvantages in the current resource scheduling algorithms. It is an urgent need to figure out how to predict the resource state more correctly in order to direct the resource scheduling algorithm, and how to improve the resource scheduling algorithm in order to get a better scheduling result.This paper firstly introduce the conception of Grid, summarize the status quo and the development trend of Grid in recent years, and expatiate the importance of the resource scheduling in the research of Grid Computing. Then a Grid File Sharing Model (FsvGrid) based on virtual organizations is proposed in order to create a platform for management and sharing of file resources in Grid environment. In addition, we thoroughly analyze and compare various static and dynamic scheduling algorithms. Based on this, a resource prediction model is proposed. The prediction mechanism can predict the periodicity and the abnormality of the resource. It modifies the model of the prediction and then gives better predictions by distinguishing between the stability and instability of the states, periodical time and abnormal time. Through the result of the prediction, an improved genetic algorithm can be used to schedule the resources. In this algorithm, we consider both the quota of tasks distributed and the finish time of tasks, and design a good mechanism of cross-over and choosing in order to keep individuals which have small finish time or parts of resources well distributed. This algorithm has better efficiency and constringency. At last we simulate our algorithm with the aid of GridSim toolkit and verify that our algorithm is reasonable and efficient.