云存储中面向访问任务的小文件合并与预取策略  

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

作　　者：王涛[1] 姚世红[1] 徐正全[1] 熊炼[1] 

机构地区：[1]武汉大学测绘遥感信息工程国家重点实验室,武汉市珞喻路129号430079

出　　处：《武汉大学学报（信息科学版）》2013年第12期1504-1508,共5页Geomatics and Information Science of Wuhan University

基　　金：国家973计划资助项目(2011CB302306);国家自然科学基金资助项目(41271398)

摘　　要：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。Aiming to improve the small file problem of a general distributed file system in cloud storage, a file merging and prefetching strategy based on a users＇ access task is proposed 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 selects prefetching files based the transition probability of the tasks. Efficiency model analysis and experimental results of a digital city prototype system based on HDFS all show that the proposed strategy has a high prefetching hit ratio and can effectively reduce the metadata servers＇ load and the response delay for users＇ request.

关 键 词：分布式文件系统 概率潜语义分析 小文件 访问任务 合并与预取 

分 类 号：P208[天文地球—地图制图学与地理信息工程]