[1]屠雪真,黄震江.一种海量小文件对象存储优化方案[J].计算机技术与发展,2019,29(08):31-36.[doi:10. 3969 / j. issn. 1673-629X. 2019. 08. 006]
 TU Xue-zhen,HUANG Zhen-jiang.A Mass Small File Storage Optimization Scheme Based on Object File System[J].,2019,29(08):31-36.[doi:10. 3969 / j. issn. 1673-629X. 2019. 08. 006]
点击复制

一种海量小文件对象存储优化方案()
分享到:

《计算机技术与发展》[ISSN:1006-6977/CN:61-1281/TN]

卷:
29
期数:
2019年08期
页码:
31-36
栏目:
智能、算法、系统工程
出版日期:
2019-08-10

文章信息/Info

Title:
A Mass Small File Storage Optimization Scheme Based on Object File System
文章编号:
1673-629X(2019)08-0031-06
作者:
屠雪真1 ;?黄震江2
1. 河南大学 计算机与信息工程学院,河南 开封 475001; 2. 南京中兴新软件公司,江苏 南京 210012
Author(s):
TU Xue-zhen1 ;?HUANG Zhen-jiang2
1. School of Computer and Information Engineering,Henan University,Kaifeng 475001,China; 2. ZTE Nanjing Corporation,Nanjing 210012,China
关键词:
对象文件系统;?小文件;?元数据;?聚合结构;?查表索引;?预读
Keywords:
object file system;?small file;?meta data;?aggregate structure;?lookup table index;?read ahead
分类号:
TP31
DOI:
10. 3969 / j. issn. 1673-629X. 2019. 08. 006
摘要:
在海量小文件存储场景下,传统分布式文件系统存在元数据服务器性能瓶颈、存储空间浪费严重、磁盘 I/O 效率低等问题。 业界主要采用小文件聚合的方法解决这个问题,但现有研究依赖于从聚合结构到小文件的二次映射和查表检索等传统方法。 文中提出一种基于对象文件系统的海量小文件优化方案,根据局部性特征将小文件聚合为文件组,使用算法直接进行对象数据存储位置的分布与定位,将低效的查表检索方式改变为高效快捷的“计算检索冶方式,这更加适合大规模分布式系统的设计;在客户端采用小文件数据大粒度预读技术,聚合小粒度 I/O 为大粒度 I/O,提升了磁盘访问效率,使用页面热缓存和温缓存两级队列管理及识别热数据,并利用文件的局部性特征提升缓存命中率。 实验结果表明,在海量小文件随机读写场景下性能提升 50%左右。
Abstract:
In the case of massive small file storage,traditional distributed file system has problems such as metadata server performance bottleneck,storage space waste and low disk?
I/O efficiency. The small file aggregation is mainly used to solve this problem in the industry,but the existing research relies on traditional methods such as secondary mapping and table lookup retrieval from aggregation structure to small files. We propose a massive small file storage optimization scheme based on object file system. The small files are aggregated into file groups according to local features,and the distribution and location of object data storage location are directly carried out by the algorithm, which changes the inefficient look-up table search to an efficient and fast “computation search”. The method is more suitable for the design of large-scale distributed system. In the client,the large-grained pre-ahead technology of small-file data is adopted to aggregate small-grained I/O into large-grained I/O,which improves disk access efficiency. Queue management at both page hot cache and hot cache levels is used to identify hot data,and cache hit ratio is improved by utilizing local feature of files. Experiment shows that the performance is improved by about 50% in the random reading and writing of small files.

相似文献/References:

[1]王全民,张程,赵小桐,等. 一种Hadoop小文件存储优化方案[J].计算机技术与发展,2016,26(11):41.
 WANG Quan-min,ZHANG Cheng,ZHAO Xiao-tong,et al. A Small Hadoop File Storage Optimization Scheme[J].,2016,26(08):41.

更新日期/Last Update: 2019-08-10