一种结合SSD特征的分布式文件系统元数据优化技术  

Metadata Optimization Method for Distributed File System Based on SSD Feature

作　　者：罗圣美[1,2] 陆游游[1] 秦雄军 杨洪章[2] 张佳程[1] 舒继武[1] LUO Sheng-mei;LU You-you;QIN Xiong-jun;YANG Hong-zhang;ZHANG Jia-cheng;SHU Ji-wu(Department of Computer Science and Technology ,Tsinghua University ,Beijing 100084 ,China;ZTE Corporation, Nanjing 210012, China)

机构地区：[1]清华大学计算机科学与技术系,北京100084 [2]中兴通讯股份有限公司,南京210012

出　　处：《小型微型计算机系统》2018年第5期930-937,共8页Journal of Chinese Computer Systems

基　　金：国家自然科学基金项目(61433008;61502266)资助;北京市科技计划课题项目(D151100000815003)资助;中国博士后科学基金项目(2016T90094;2015M580098)资助;深圳市科技创新委员会科技应用示范项目资金(SF20170036)资助;中兴通讯产学研合作项目(PJ151220124237)资助

摘　　要：分布式文件系统的元数据性能是制约系统整体性能的关键瓶颈.尽管固态盘(Solid State Drive,SSD)提供高速的数据访问,但是由于元数据呈现粒度小、更新频繁的特征,SSD的性能表现仍然较差,同时导致寿命损耗加速.基于SSD存储介质的写入特性,提出了面向分布式文件系统元数据的数据管理机制和更新方法,包括元数据内存页面的重新组织和管理、多次变化数据的迭代更新、元数据写入方式的进一步优化等.所提方法减少了元数据更新的写入频次和实际写入量,减少了随机写操作,提高了元数据写入性能.The performance of metadata is a key design issue in a distributed file system. Although SSDs provide high-speed data access,the performance and endurance of SSDs are still poor due to the small granularity of metadata and the frequent updating. Based on the writing characteristics of SSD storage media,this paper puts forward the data management mechanisms and updating methods for metadata of distributed file systems. Those mechanisms contain the reorganization and management of metadata pages,iterative updating of multiple changed data,metadata writing mode optimization and so on. The proposed methods reduce the write frequency,the actual writesize of the metadata updates and the number of random write operations,and thereby improve the metadata write performance.

关 键 词：固态盘 分布式文件系统 元数据 数据管理 

分 类 号：TP333[自动化与计算机技术—计算机系统结构]