摘要： 在分布式文件系统中,对元数据的预取能够减少元数据服务器的请求响应延迟时间。现有的元数据预取策略大多基于元数据的历史请求序列,并未考虑文件的起源信息。为此,提出一种基于起源信息窗口的元数据预取策略,通过分析进程行为与元数据请求的关联性,提取起源信息窗口,统计元数据文件之间的关联度,生成关联规则哈希表,进行更激进的元数据预取。实验结果表明,与传统的最近最少使用算法和基于权重有向图的元数据预取算法相比,该策略的Cache命中率分别提高49%和7%。与Nexus算法相比,能有效减少内存开销,提升关联规则的查询效率。

关键词: 起源信息, 元数据, 预取, 文件关联, 分布式存储

Abstract: In distributed file systems,metadata prefetching can reduce the response latency of metadata server.Existing metadata prefetching strategies prefetch metadata according to its past file access pattern,without considering the correlations between process and the corresponding files,such as file’s provenance.A metadata prefetching strategy based on provenance information is proposed for distributed storage systems.The strategy extracts provenance information windows,calculates the correlation degree of any two metadata files after analyzing the correlations between process and metadata request.It generates the correlation rules hash table and employs aggressive metadata prefetching.Experimental results show that Cache hit ratio of proposed strategy is up to 49% and 7% respectively.Additionally,the proposed strategy performs more effectively with less memory overhead than Nexus algorithm.

Key words: provenance information, metadata, prefetching, file correlation, distributed storage

中图分类号: 

• TP311