Study On The Strategy Of Dynamic Replication And Replica Selection In Data Grid

Grid is a kind of seamless, integrated resource sharing and collaborative environment. In order to achieve the sharing of computing resources, storage resources, data resources, knowledge resources and expert resources, the Grid connects dispersed computers, storage equipment, and scientific equipment together with the network, and integrates into a huge virtual super computer. Generally speaking, the size of Grid is relatively large, and its essence is distributed, heterogeneous and dynamic. The Grid implements collaborative work and sharing of resources between virtual organizations, providing reliable quality of service. As an important branch of Grid, the Data Grid which mainly concentrates on data-intensive applications, links network nodes locating in different geographical positions through the Grid infrastructure. The goal of Data Grid is to establish an integrative storage, management, access, and transmission of the massive data in the distributed heterogeneous environment, as well as the framework and environment of related services to realize the sharing of data and resources effectively. Data replication is one of key techniques in the Data Grid, whose goal is to obtain better data access performance. By placing replica on appropriate nodes, replication technology can supply the users with local data replica which can be accessed and processed fast, avoiding a large number of long-distance data transmissions, thus can reduce access latency and bandwidth consumption greatly, and can improve the system reliability.This thesis analyses the present research situation of Data Grid, and summarizes the deficiencies of current replication strategy. According to the specific circumstances of the educational resource gird, it constructs a replicated catalog model of inter domain and inner domain, and designs a dynamic replication strategy based on above metioned. The main work and innovation of this thesis are listed as follows:1. By investigating a lot of literatures, this thesis compared several data replication strategies widely used at present. It summarizes the deficiencies of the replication management, and analyses the improved methods simply.2. According to the characteristics of educational grid, a reasonable model of the replicated catalog is established. In this model, it divides grid nodes into many domains, unifies the name space of nodes, employs a double-layer catalog in combination with central catalog and middle catalog, and discusses how to execute the replica location and the replica consistency.3. The center nodes of each domain connect each other through the P2P network, thus the number of nodes which are managed by the central ones is fewer, and the single node failure problem can be improved effectively. In a single domain, it uses Lightweight Directory Access Protocol - LDAP as the protocol to access catalog information, and applies a mixed topology construction combines with tree and ring which can lighten the burden on the central node, and raise the efficiency of catalog information index and data transmission.4. It defines the load of grid nodes strictly, designs a load-based dynamic replication strategy, which focuses on the replica location and replica selection, and controls the number of replica with replica replacement strategy. The replica location strategy can place the replica accurately on the overload nodes, and has a strong dynamic characteristic to can adapt to the high changes of data request. Moreover, the strategy of replica selection can improve the response and transmission speed of data request.5. This thesis compares several grid simulators, and analyses the reason for choosing the OptorSim. It simulates the replication strategy mentioned in this thesis, and verifies the accuracy and performance of the strategy. Experimental results show that the proposed algorithm and strategy is feasible, and can enhance the speed of data transmission.