一种细粒度高效多版本文件系统

An Efficient Fine Granularity Multi-Version File System

提出了基于快照的细粒度版本技术,能够克服已有多版本文件系统无法仅对系统局部目录或文件保留版本的缺点,增加了系统的灵活性;提出了版本空间的反向继承寻径,使用名字与版本独立的检索方案,可以充分利用版本间的相关性,优化数据物理布局,建立版本间的层级结构,既便于管理,又提高系统性能;设计了分别针对目录版本和文件版本的快速索引结构.评测结果表明,THVFS的历史数据访问性能较著名的多版本文件系统ext3cow提高了34.4%;Trace实验中,相对于ext3,THVFS的读性能提高了12%,同时,在每72分钟生成一次快照的高频率下,维护所有历史版本仅需要80%的额外空间.

A snapshot-based fine granularity versioning technique is presented to retain history data only for a single directory or a single file, and bring flexibility to multi-version file systems. Adopting the strategy to search in name space and version space separately, this paper also presents backward inheriting path-finding mechanism in version space. This mechanism is beneficial to the performance and management, because it can utilize the coupling relationship between versions to optimize the data layout of versions and build hierarchy in version space to accelerate the path-finding procedure. In addition, fast index structures for directory versions and file versions are designed. This prototype file system--THVFS can achieve both good performance and high availability with these technologies mentioned above. The experimental results show that the average time of searching old versions in THVFS was reduced by 34.4% than in ext3cow, the famous multi-version file system. In the trace experiment, the average read response time in THVFS was 12% less than in ext3, and only 80% extra space was needed to retain all history data when snapshots are taken every 72 minutes in THVFS.