Design And Implementation Of Resource Management And Allocation System Based On Grid Computing

With the popularity of the Internet and the significantly cost reduction of high-speed networks, also with the change of the traditional method of calculation and the use of computers, grid computing has become an important trend in the development of super computing. Since resource characteristics of the grid computing environment make the resource management of grid computing to be a rather challenging problem, the complex computing resources in wide area will greatly influence the performance. Grid resource management system and the resource management system of traditional distributed computing or cluster computing are similar, it should have at least three functions: the requested operations management, resource management, resource allocation and scheduling. That also means the users could submit the resource request to the network system by the requested operations management, Identify and monitor the status of network resources, collection the resources information of running job and others information, manage the cluster resource, allocate the resource according to the scheduling mode.Based on the above requirements, resource management and resource allocation are described and explained in the dissertation. In the terms of resource management, the concept of resource groups is used to manage compute nodes. A variety of resource allocation are used so that grid computing resources are better used, it also provides a basis about the effective management and use for grid computing resources. In the terms of resource allocation, the concept of consumer tree is used, the resource are allocated to the resource according to the consumer tree.The common problems in resource management and allocation such as preventing deadlock, scheduling problems, load balancing and other issues are analyzed and the solutions are given in the dissertation, so that the system could manage and allocate the resource efficiently and stably. Atomic transaction based resource allocation algorithm is used to prevent the possible occur of resource allocation deadlock. In order to absorb the advantages of centralized and distributed scheduling for solving the bottleneck for cluster computing and scheduling, a three-level model is selected as the system scheduling mode. Load balancing of computing nodes is achieved by defining the three states account for compute nodes. A mixed service queue based polling system is used. The polling system could deal with larger number of resource requests, at the same time, handle requests for different priority depending on the application of the request priority.