The Research Of Replica Selection And Consistency For Grid

The Data Grid provides a high-performance and high-capacity parallel distributed computing platform to store and manage the massive data in the distributed heterogeneous WAN environment. Replication is introduced into Data Grid in order to reduce access latency and bandwidth consumption and enhance the reliability of data. Though Replication can improve the overall performance of the grid system, the management of the replicas still exists. Replica selection and consistency management is the core of replica management. They influence the property and performance of the Data Grid.The client is the center in the weight based replica selection mechanism. It selects bandwidth, the speed of disk I/O and the load of the replica node as the selection parameters, the client gives the weight of the parameters. Through location module to locate the replicas, information service module to get the value of the parameters, selection module to compute the whole weight and then select the best replica for client.In improved adaptable replica consistency update strategy, replicas are separated into three categories: source file, master replica and secondary replica. Different replica has different update strategy. Up-to-date modification is for source file and master replica; update based on access time is for secondary replica. At the same time, locking mechanism can resolve the problem of synchronization.Through the test of the system, the results indicate that the weight based replica selection mechanism keep balance between replicas and reduce the average access time efficiently; the improved adaptable replica consistency update strategy can reduce the total access time and the time of replication.