云存储中面向访问任务的小文件合并与预取策略

针对云存储中通用分布式文件系统的小文件问题,改进概率潜语义分析(PLSA)模型,提出了一种面向用户访问任务的小文件合并与预取策略。该策略分析用户的访问任务、系统应用和访问文件之间的关系,根据任务合并小文件,并基于任务的转移概率预取文件。对建立的效率模型的分析和基于HDFS的数字城市原型系统实验结果都表明,此策略有较高的预取命中率,可以有效减少元数据服务器的负载和用户请求响应时延。

A Small File Merging and Prefetching Strategy Basedon Access Task in Cloud Storage

Aiming to improve the small file problem of a general distributed file system incloud storage,a file merging and prefetching strategy based on a users’access task is pro-posed that improves the PLSA model.Analyzing the relationships among the access tasks,applications,and access files the strategy merge small files on the basis of tasks and selectsprefetching files based the transition probability of the tasks.Efficiency model analysis andexperimental results of a digital city prototype system based on HDFS all show that the pro-posed strategy has a high prefetching hit ratio and can effectively reduce the metadata serv-ers’load and the response delay for users’request.