快速瀑布模型动态副本创建策略研究

Research of Fast Cascading Dynamic Replication Strategy for Data Grid

在高性能数据网格中,动态副本创建策略能使用户访问远程大数据文件时有效的减少带宽消耗和访问延迟。提出了多层数据网格环境下一种新的动态副本创建策略——快速瀑布模型算法。OptorSim网格仿真器也被修改用以更好的适应多层数据网格架构和研究这种算法的性能。仿真结果表明:对比Fast Spread,LRU,LFU,Economy-Bin和Economy-Zipf五种算法,快速瀑布模型算法能在随机访问模式下有效地减少数据访问的平均响应时间,并且其优势随着用户提交作业数量的增加而增大。

Dynamic replication can be used to reduce bandwidth consumption and access latency in high performance data grids where users require remote access to large files. A new dynamic replication algorithm, Fast Cascading, was proposed for the multitier Data Grid. Simulator OptorSim was modified to study the performance of the replication algorithm. The simulation results show that this algorithm can reduce the average response time of data access greatly for random access pattern compared to Fast Spread, LRU, LFU, Economy-Bin and Economy-Zipf dynamic replication strategy and the gain increases with the number of jobs submitted.