【作者】 张翔；

【导师】 冯丹；

【作者基本信息】 华中科技大学 ， 计算机系统结构， 2007， 硕士

【摘要】 随着网络技术的飞速发展和用户对大规模存储系统日益增长的需求,网络存储系统逐渐成为存储系统领域的主流。对象存储一改传统的块或文件接口,采用富有表现力的对象作为存储接口,克服了块接口与文件接口的缺陷。HOSS(Huge Object Based Storage System)是基于Linux平台的对象存储系统,由元数据服务器、客户节点和基于对象存储设备三部分构成。在现实中,有很多重要的应用是基于Windows系统的,比如GIS应用,而现有的HOSS客户端,以及已知的其他对象存储系统的客户端,都是基于Linux系统的,不能够满足这些系统的需要。为此设计并实现了基于NT系统的用户文件系统HIFS(HUST Installable File System)。HIFS系统是一个过滤器型文件系统,它位于文件系统驱动之上,能够截获用户文件操作,并重定向到后端的对象存储系统上,用户在Windows下能像操作一个本地磁盘那样来操作HOSS系统,获取存储空间和文件的共享服务。详细讲述了HIFS系统的I/O操作部分的开发与实现,并使用聚合写等技术提高系统I/O的性能。针对现有的OBS系统客户端之间不能有效共享数据的问题,提出在客户端文件系统内引入对等技术的方案,并设计了一种一致性维持机制的方案,即使用一种双Cache的结构来维护文件对象的一致性;使用一种可订制的一致性级别的机制来提高系统的I/O性能,避免了锁机制带来的瓶颈效应。更多还原

【Abstract】 As the development of the network technique and the requirement of the storage system’s size grow up, network storage system became more and more important at the storage system domain. Object based storage had changed the interface of the traditional block or file based storage systems. Object based storage system used the expressive object as the storage interface, had solved the limitation of the block or file interface. Object based system consist by Meta data sever, client and object based storage device.HUST storage lab had developed an object based storage system called HOSS. This storage system was built on Linux platform. Actually there are so many important applications are design for Windows just as the GIS system. But now days all the clients are based on Linux operation system, can not be easy used in Windows.So, IFS group design and implement a client file system driver (FSD), called HUST Installable File System (HIFS), in Windows NT environment. HIFS is a filter driver, which is higher-level of FSD. File I/O from user mode are intercepted by this driver, and then redirected to the Object-base device. The client of HOSS system can be operated as a normal local disk. From this logical disk, User acquisition the sharing service of storage space and file. This paper figures on the development the I/O operation model of HIFS. In this section, several algorithms are present to optimize the system performance, such as aggregate-write and so on.All of the Object-based storage system mentioned above can’t share the data among the clients. In this paper, based on P2P technique show an approach to maintain the consistency of client shared files. A Dual-Cache is designed to help us achieve this goal. Furthermore, a tunable mechanism of consistency level is used to resolve the bottleneck by using lock and improve the I/O performance of system.更多还原