Research On Replica-management Scheme For Data Grid

With the use of data grid becoming more and more widespread, the replica of data grid turns increasingly important. However, as the limitation of replica delay, many present projects could not be achieved through data grid. As a result, the replication issue becomes one of the research hotspot.Because of the different application mission of data grid,the requirements of data grid remote transmission delay and replica consistency are different. When data grid is applied to transmit data stream, the delay between nodes is not very important. when data grid is applied to transmit big packet, the delay is very important. In this paper, according to the different application tasks of data grid, we proposed two replica replacement schemes to meet the respective needs of tasks.Based on one by one approach and the abort and retransfers approach, this paper presents a"front-end paralleling"schema at first. When a replica server has completed its assigned tasks, it queries other replica servers to find the one which has transmitted the least block. Data grid allocates some blocks to the quicker server proportionally. Two servers finish the task cooperatively at the same time which reduces the remote transmission delay.Furthermore, when user needs both complete and fragment replica, the paper presents the strategies of replica creation, replica replacement and replica selection of data grid. This proposal meets different data types, different node delay and server performance. It can improve the system performance. Through theoretical analysis and simulation of the"front-end paralleling"schema and the new replica creation strategy, it proved that this program has better performance if the performance gap between replica servers is serious.