Research On Replica Management Strategies In Education Resource Grid

Grid technology enables the sharing and collaboration of wide variety of resources. In order to organize and manage discretely distributed education resources and achieve the sharing of education resources, education resource grid appears. Education resource grid aims at education resources sharing. In other words, users can share education resources on the premise of protecting intellectual property rights, and repetitive blind exploitation of softwares can be avoided in effect. Replica Management is the basis of grid resources sharing, and it is one of the most important standards to evaluate the performance of grid system. To make a target of improving the efficiency and stability of resource access, replica management strategies in education resource grid are researched thoroughly in the context of National Science Foundation of China under Grant.Based on previous researches, major research works and innovative points in this thesis are:Firstly, this thesis analyses the major problems that education resources sharing exist, summarizes the characteristics and developing prospects of education resource grid, and gives the definition of that education resources grid. According the physical network characteristics of education resource grid, its network architecture is defined. In this network architecture, the city domains are composed of nodes which have short access delay with each other or are in an adjacent site. Based the characteristics of education resource, the model of replica management in education resource grid is proposed, and the main problem of replica management is given in mathematical formulation, which lay the foundation for the follow-up researches.Secondly, according the characteristics of the physical network in education resource grid, an improved routing algorithm, DC-CAN, is given on the basis of the CAN algorithm. In the DC-CAN algorithm, the nodes in the same city domain are allocated to a continuous CAN region, which avoids many hops between two different city domains in the routing process. Therefore, The DC-CAN can adapt the physical network of education resource grid effectively. Then, a replica location strategy on the basis of DC-CAN is proposed, which reduces the average delay of routing on the premise of load balancing.Thirdly, the forecast of access traffic based replica placement strategy is given according to the network architecture. In the strategy, the set of hot files will be picked up by forecasting the files access traffic in next management period, and the quality of the hot files access will be assured by managing the replica of the hot files. Therefore, the stability and the average speed of accessing resources are improved. At the end the thesis gives two forecasting methods of accessing traffic.Finally, it defines each function module of the replica management in education resource as service interfaces, and proposes a grid service scheduling architecture based on economic model (GSAEM). Then, in order to give a precise mathematical model to the price regulation of grid service under the economic model, the marginal principle based price regulation is proposed. It deduces the best prices of grid service under all conditions in various resource constraints by marginal principle analyzing, and therefore it optimizes the performance of service-oriented scheduling.The algorithm in grid environments needs simulating in order to validate the accuracy and high performance. First, the implement of date accessing in replica management is programmed in the thesis, which gets better experimental results by comparing with traditional file-sharing model. Second, according to some rules, the network architecture topology of education resource grid is generated by using the tools of BRITE. The thesis designs the simulation of DC-CAN, and the results of the simulation confirm that the algorithm has short average delay in replica location and better performance in load balancing. Last, it simulates the algorithm of the replica management service scheduling, and demonstrates that marginal principle based price regulation mechanism achieve superiority in scheduling.