基于pNFS的小文件间数据预读机制研究

A Research of Data Readahead Mechanism Between Small-Files Based on pNFS

随着信息技术飞速发展,小文件的数量越来越多.高效管理海量小文件、提供低延时的小文件访问服务,是并行网络文件系统的研究热点.小文件数据读取访问中,由于数据读取粒度小并且不同小文件之间的数据访问空间连续性差,难以发挥数据存储设备的大粒度顺序访问的性能优势,导致小文件的访问性能远远低于大文件的访问性能.在扩展只读目录授权机制框架下,提出一种小文件间数据预读机制,充分发挥了数据存储设备大粒度顺序访问性能的优势,大幅降低了小文件数据读取访问延迟.典型应用环境实测表明,对比pNFS系统,小文件数据顺序读取访问性能最高可提升8.92倍,随机读取性能最高可提升11.82倍.

With the rapid development of information technology,there will be more and more small files.How to manage massive small-file efficiently and provide low-latency service is a hot topic in pNFS.When reading access to the data of massive small files,because data granularity is small,and the continuity of data access space is bad between small files,massive small file access performance is far lower than large file access performance,and the performance advantage of Data storage devices large-grained sequential access is unable to show.This paper presents a data Readahead mechanism between small-files under extended read-only dir delegation.This mechanism can slashing small file read access latency,and as well display data storage devices large-grained sequential access performance advantage.The effectiveness of this paper is proved with intensive experiments:when accessing massive small-file,compared with pNFS,the performance of sequential read access is 892%higher,and the random read access performance is 1182% higher.