Research And Application Of Resource Management And Scheduling Method In Grid

Grid is a new field of research rapid developing in recent years, the emergence of which represents that network technology has entered a new epoch. Grid is the essence of sharing and collaboration, that is, controllable and equitable use and sharing of resources in the dynamic, scalable virtual organization. How to make full use of the shared resources is the key factor of determining the performance of grid system, which could be dealt with by resource management and job scheduling. The resource management and job scheduling is one of the center and key elements of grid system. Therefore, in order to improve grid computing efficiency, we must continue to strengthen the study of resource management and job scheduling.Combining theory with practice, this dissertation systematically studies the technologies of grid resource management and schedule. The major research work and contributions in this dissertation are as follows.This thesis systematically analyzes the conception, characteristics, classification and architecture of the grid, summarizes and compares the characteristics of current grid projects. With the careful analysis of the characteristics of grid resources, summarizes and sums up the goal, characteristics and component of grid resource management and scheduling and corresponding structure model.In order to avoid the herd behavior easily raised in grid resource management and scheduling, based on the analysis of d- random selection method and non-uniformity random selection algorithm, combined with the characteristics of grid environment, we propose an improved random selection algorithm—twice random selection algorithm, which integrates the idea of d- random selection method and non-uniformity random selection algorithm, and makes up the deficiency that the non-uniformity random selection algorithm does not adapt to grid scheduling.With the theoretical analysis and practical validation, we prove that the twice random selection algorithm is indeed superior to d- random selection algorithm in load balancing, when the load information of system is not true and authentic. Therefore, the twice random selection algorithm can deal with the herd behavior better.In addition, based on recent grid technologies, with using the GRAS mechanism in SimGrid, we design and implement a system prototype of grid based on twice random selection algorithm, and found by simulation and emluator that on the basis of the prototype, the twice random selection algorithm compared to the d- random selection algorithm achieves the better load balance without bringing about too much additional overhead.The study in this dissertation is significant in theory and practice to the research of resource management and job scheduling of grid.