碧海文件系统元数据节点的优化与分布式实现  

Optimization and Distributed Implementation of the Metadata Node in BlueOcean File System

作　　者：李永泽[1] 

机构地区：[1]上海交通大学信息安全工程学院,上海200240

出　　处：《小型微型计算机系统》2014年第9期1989-1993,共5页Journal of Chinese Computer Systems

基　　金：上海市科学技术委员会基金重点项目(10DZ1500200)资助

摘　　要：碧海文件系统是横向扩展的大规模分布式文件系统.本文详细描述了碧海文件系统元数据节点的优化与分割方案:通过对目录树进行子树分割,将元数据布局在不同的节点之上;实现了监控节点,能够仲裁元数据节点的状态,收集并分发子树信息,增减元数据节点集群的规模;优化了多个数据节点与元数据节点的通信模型,避免了重命名操作造成文件数据的移动;设计了分布式节点的事务处理方式,使元数据信息可以高可靠地在不同节点之间移动.测试表明,通过以上的改进,碧海文件系统的扩展性与并发性能有了显著的提升,并且对于重命名之类跨节点的操作,也能较好的支持.Abstraet：BlueOcean system is a scale-out massive distributed file system. This paper elaborates the optimization and partitioning scheme of the metadata node in BlueOcean file system： Splitting the dentries into sub-trees arms the system with the capability of distributing metadata on different nodes; Implementing the monitor nodes makes tasks of arbitrating metadata nodes＇ status, collecting and handing out information on sub-trees, and changing the size of the metadata node cluster dynamically feasible; Optimizing the model with which multiple data nodes interact with multiple metadata nodes, avoids the gigantic file data movement resulting from the rename operation; designing the distributed transaction processing between separate nodes, contributes to highly reliable migrations of metadata information. The results of corresponding experiments illustrate there is a great gain in the scalability and concurrency of the BlueOcean file system through the endeavors mentioned above, what＇s equally important the system performs well with the operations such as rename which require collaborations between individual metadata node. Key words： management of metadata; sub-trees partition; transaction processing; scalability; distributed file system

关 键 词：元数据管理 子树分割 事务处理 可扩展性 分布式文件系统 

分 类 号：TP311[自动化与计算机技术—计算机软件与理论]