Resource Management Strategies And Corresponding Algorithms In Data Grid

Grid is generalized and large-scale distributed network system that can scale to Internet size environments with resources distributed across multiple organizations and administrative domains. Data Grid is one of the branches of Grid that provides a mechanism for effective distributed resources sharing and transparent remote access to heterogeneous data on wide-area network. There still are challenging key issues in designing a resource management system in Data Grid.Targeting on the existing problems and defaults in resource organization and location in Data Grid, in this dissertation, we review the present condition of investigation in-depth, apply Small World theory to resource organization field, and present a query pattern of resource location based on semantic group. In order to improve the security of resource locating, trust community is used in resource locating mechanism. Replication in Data Grid brings a series of replication management problems that greatly affect the performance of Data Grid system and therefore replica management, a part of resource management, becomes a hot topic in Data Grid. The thesis provides a replica creation strategy and algorithm as well as a replica selection strategy.Theoretical analysis and simulation results show that our methods are feasible and efficient. Thus, our research on resource management in Data Grid environment is an exploring of important realistic and practical significance.The main contributions of the dissertation are summarized as follows:â‘ Based on Kleinberg Small World model, we propose an efficient decentralized construction approach SSWCP, which uses partial view instead of global knowledge of network to reduce network diameter. SSWCP is then utilized in a new resource organization approach SWLS in which semantically related nodes form layered semantic groups. Queries are routed to the appropriate semantic groups, increasing the chances for quickly finding matching resources, and reducing the search load on nodes that have unrelated resources. As shown by theoretical analysis and experiment, scalability and load balance are improved compared with system unprocessed.â‘¡Resource locating algorithms are established based on SWLS. Meanwhile, how to securely locate resources is an important factor affecting the performance of Grid system. We propose a novel method in which trust relations implicitly existed among network nodes are utilized to construct a secure and available resource trust community (SARTC) for resource discovery. Through analysis, the proposed solution that combines SWLS with SARTC is scalable, secure and efficient.â‘¢Data Grid provides geographically distributed data resources, however, ensuring efficient and fast access to such huge and widely distributed data, however, is hindered by the high latency of the Internet. Replication is an effective approach to address these problems. We introduce a system model of replica creation and mainly discuss an adaptive replica placement strategy RGRS,which depends dynamically on current conditions, such as user access, network topology, and so on, offering improved scalability of the overall system. Through analysis and experiments, the results demonstrate that the strategy is self-adaptive and scalable, and improves the performance of the data access in Data Grid.â‘£When different nodes hold replicas, there are significant benefits for selecting the best replica. We propose a new replica selection algorithm based on prediction. When we analyze the factors infecting replica selection, the gray system theory is employed to help predicting the data response time on the basis the GM(1,1) gray dynamic model; meanwhile the Markov chain is used to achieve the state transition probability matrix for predicting the reliability of replicas in the form of probability by the system state classification. The simulation results show that in our replica selection model, the approach is valid and helpful to selection decision, and is able to achieve load balance between nodes holding replicas in our replica selection model.Thus the dissertation may provide new clues for future research work.