节点文献

对象存储原型系统设计及相关实现

The Design and Implementation of Object-Based Storage Prototype System

【作者】 史伟

【导师】 冯丹;

【作者基本信息】 华中科技大学 , 计算机系统结构, 2006, 硕士

【摘要】 数字以强大的信息表达能力以及单一的处理、传输和存储方式,融合了整个信息技术。半个世纪以来,作为数字信息载体的存储技术得到飞速发展。不断增长的存储需求和管理成本催生了基于对象的存储技术,而“对象”也有望成为下一代存储技术的标准接口。在分析当前流行的网络存储体系结构及存储协议的基础上,对基于对象存储技术以及T10的SCSI OSD(Object-Based Storage Device Commands)标准作了深入研究。“对象”是传统块接口和文件接口的折中,基于对象存储系统在I/O性能、跨平台、可扩展性以及安全性等方面都表现不错。实现了一个符合T10 SCSI OSD标准的对象存储原型系统,包括对象存储设备和客户端。客户端的SCSI对象设备驱动是一个Linux SCSI上层驱动,基于Linux块设备子系统实现,用来管理所有检测到的OSD设备。iSCSI启动设备是Linux SCSI协议栈的底层驱动,为客户端提供通过IP网络访问iSCSI目标设备的iSCSI通路。iSCSI目标设备实现iSCSI传输协议的Target部分。对象存储服务模块负责管理物理存储介质和处理OSD命令,以上模块均在Linux内核空间实现。测试并分析了基于对象存储原型系统的性能,得出的结论是:通过在对象存储原型系统引入聚合读/写机制可以大大提升系统的I/O带宽。

【Abstract】 By virtue of its powerful capability in information presentation and uniform way to process, transmit and store, digit has unified information technology. Half a century has witnessed the development of storage technology. Ever increasing storage demands and management costs has given birth to Object-Based Storage technology and“object”is on the verge of becoming the next standard storage interface.On the basis of the analysis of the current prevailing network storage architecture and the storage protocols, Object-Based Storage technology and T10 SCSI OSD standard are studied insightfully. Objects can be regarded as convergence of files and blocks. Object-Based Storage System proved advantageous in performance, cross-platform, scalability and security.An object-based storage prototype system based on T10 SCSI OSD standard was designed, which involves object-based storage device and client. Client’s SCSI object driver is realized as a SCSI upper level device driver, whose main function is to manage all detected OSD devices; iSCSI initiator is a SCSI low level driver providing iSCSI transport to access remote iSCSI targets over IP networks; iSCSI target implements the target side of iSCSI transport protocol; Object-based storage server manages the physical storage media and processes OSD commands. All are realized as Linux kernel mudules.In the end, according to the performance test and analysis, we conclude that by introducing mechanism of aggregated read/wirte I/O throughput of object-based storage prototype system is drastically improved.

  • 【分类号】TP311.52
  • 【被引频次】1
  • 【下载频次】109
节点文献中: 

本文链接的文献网络图示:

本文的引文网络