Research On Key Technologies Of Replica Management In Data Grid

A large amount of computing resources and storage space were provided by data grid to deal with large-scale data-intensive applications, and the sharing of various resources was realized in the worldwide. Because of the limitation of available resources and the production of large-scale workload in grid, the effective use of grid resources has become a major challenge. Data replication can reduce the bandwidth consumption, improve the fault-tolerant ability and enhance the extensibility. Therefore, the data replication technology has become an important issue of grid research. In view of the poor efficiency of the current replica management strategy, a new one based on the response time and the copy cost was proposed in the paper through in-depth research on a variety of replica management strategies.Firstly, the currently existing grid topology in the replica management strategy was analyzed in the article, according to their advantages and disadvantages, a 3-level bandwidth hierarchical structure was chosen for the basic network structure of the research. Aiming at the limitations that a single replica selection factor was only considered by the replica selection algorithm in the current replica management strategy, and considering the waiting time of the replica request queue in the storage and the data transfer time, a replica selection algorithm based on the replica response time was presented to select the best replica. At the same time, a replica placement algorithm based on replica demand was put forward.Secondly, when the current traditional replica replacement model and replica replacement strategy had been analyzed, it’s found that the replica use frequency was only considered and the other factors were ignored, leading to inefficient system performance. In view of this situation, a replica replacement strategy based on copy cost was proposed. To avoid the unnecessary time latency and bandwidth consumption, it considered two parameters that the replica access frequency and the transfer cost to delete the file.Thirdly, by considering replica selection algorithm based on response time, replica placement algorithm based on replica demand and replica replacement algorithm based on copy cost, a dynamic replication strategy based on response time and copy cost is proposed.Finally, using the grid simulator Optor Sim as the test platform, the dynamic replication strategy based on the response time and the copy cost and dynamic hierarchical replica management strategy were simulated. The experimental data of these two management strategies was compared from the perspective of multiple performance evaluation criteria, which proves the effectiveness of the proposed strategy.