关键词: 数据网格, 副本放置, 文件流行度, 星型拓扑

Abstract: Data replication technique reduces the network latency and network bandwidth by predicting the preference of users and placing file replicas to the nearest resource node of users in advance, which has been widely used in data grid and cloud computing. Two important steps involve in data replication, file selection and replica placement. File selection is responsible for predicting the preference and selecting the most popular files as the candidates of replicas. Replica placement is responsible for placing file replicas to the most suitable node in data grid by taking the location, capacity and bandwidth of the resource node into consideration to minimize the latency and bandwidth consumption of the entire system. By redefining the file popularity, this paper improves the algorithm??s sensitivity to large files. Then this paper designs an asynchronous algorithm with distributed ideas which consists of local algorithm and global algorithm. The local algorithm sends the file access records to the global algorithm to calculate the global file popularity through the asynchronous mechanism. Then, the local algorithm selects the candidates of replica using redefined file popularity. Finally, the candidates of replica will be placed to the most suitable resource node. The simulation experimental results show that the proposed strategy can reduce the average file latency and the average bandwidth consumption to a certain extent compared with the IPFRF (improved popular file replicate first) algorithm under the access patterns of Gauss distribution and power law distribution.

Key words: data grid, replication placement, file popularity, star topology