AMS实验中的副本管理研究

Replica management research in AMS experiment

提出了一种新的副本管理策略同时结合了预测和淘汰代价因素.通过预测副本在未来时间窗内的流行程度,热点副本被保留下来可以提高平均作业响应时间.代价因素主要关注的是副本的替换代价诸如网络延时和带宽,以及副本尺寸.经过使用OptorSim仿真器对AMS国际合作实验中数据的仿真模拟,比较副本管理策略和传统的替换算法的性能,结果表明管理策略在平均工作时间和带宽消耗方面取得了很好的平衡.

It has become a big challenge to manage large volume of data efficiently produced in E-science activity.In this paper,we will present a novel replica management strategy that combines history access information and cost factors together.Applying history access information in prediction model,hot spot replica is kept down.Cost factors are most concerned about network latency,bandwidth and replica size.By simulating AMS02 data access information in OptorSim to compare replica management strategy with traditional algorithms, our management strategy achieves a good balance between mean job time and bandwidth consumption.