【作者】 孙照焱；

【导师】 冯冠平；

【作者基本信息】 清华大学 ， 仪器科学与技术， 2004， 博士

【摘要】 数字化信息的爆炸性增长推动了以光、磁介质为载体的数据存储与处理技术的研究与应用,海量数据的可靠存储与有效利用成为问题的关键。随着计算机网络的快速发展及推广应用,数据的存储模式越来越呈现出非中心化的趋势。传统的以服务器为核心的中心存储模式,数据的存储与访问高度集中,极易产生严重的I/O性能瓶颈,影响数据的有效利用,而且在系统存储容量的扩展、数据管理及防灾能力等方面均存在严重隐患。研究高性能的新型存储子系统网络应用方案十分必要。首先,论文突破传统的C/S(客户/服务器)模式,根据数据存储从单机转向网络、从集中转向分散的趋势,设计了模块式附网存储(NAS)设备作为企业内部网络的数据存储节点。模块化的体系结构使存储设备在网络中分布灵活,即插即用,具有良好的可扩展性;基于Linux平台优化的软件系统效率高、可升级、支持多种操作平台,提高了设备在网络中的适应性;远程管理方式也使设备的管理更加方便。其次,针对NAS设备固有的访问安全方面的弱点,论文研究了NAS设备的防护策略。通过分析NAS设备用户的文件访问行为,对文件请求进程中序列化信息的特点进行了研究,重点针对进程的调度及系统调用的实现进行深入研究,总结出文件请求时间的经验公式。在此基础上,引入生物免疫机制,提出由用户认证、文件权限和用户阶梯组成的多层免疫模型,从时间延迟方面分析了免疫模型的性能。实现了对用户访问行为的异常检测及自动免疫,在保证数据可用性的同时,提高了数据的安全性。在上述研究的基础上,实现了具备免疫功能的NAS文件服务器,并对设备的I/O性能进行了综合评测。结果表明,在多客户并发访问时,设备具有很高的数据吞吐性能;免疫策略的实施对文件读写性能的影响低于2%。这种基于生物免疫机制的NAS设备的研究,为超大容量数据存储子系统的网络化应用提供了一种新的方案,使非中心化数据存储节点具有高性能、高可用性、可扩展、易管理等特点。更多还原

【Abstract】 The exploded increase in volume of digital information promotes the research of storage and processing technologies for data recorded with CDs and disks. It is important for large number of data to be stored reliably and utilized effectively. With the rapid development of network, the mode of data storage presents decentralized tendency increasingly. In traditional storage mode, the server plays a role of center and highly centralizes preservation of massive data and access to them, which easily incurs the serious I/O bottleneck on the server. Moreover, scalability, manageability and survivability are not well supported in the centralized storage framework. Therefore, it is necessary to research new schemes of network application for storage subsystem.Firstly, according to the decentralized tendency of data storage, the conventional C/S mode is broken, and modularized network-attached storage (NAS) devices are designed to act as data storage nodes in LAN of enterprises. For the modularization of structure, NAS devices are easy to be distributed and utilized in network. The optimized OS based on Linux has high efficiency and scalability, and supports multiple platforms, which improves the adaptability of devices to changes in network conditions. The manageability of devices is well supported by Web-based management manner.Secondly, aiming at the weakness of access security in present NAS products, the defending strategy of NAS devices is researched. Through the analysis of the access behaviors of users in NAS devices, the characteristics of sequence information in processes of file requests are studied. Scheduling of processes and realization of system calls are carefully researched, and an experiential formula of time for file requests is summarized. Then, inspired by biological immune mechanisms, a multilevel immune model, composed of user authentication, access authority of file system and user stair, is presented to enhance the security of data stored in NAS devices. The performance of model <WP=6>is analyzed from delay-time of processes by the experiential formula. This multilevel immune model realizes the abnormity detection and automatic immunity to users’ operations, which enhances the availability and security of data in NAS devices.Based on the research mentioned above, a NAS file server with function of immunity is realized. The experimental results of I/O performance evaluation indicate that this NAS device has observably high throughput when multiple clients call it at the same time. Further more, the performance slowdown of file reading/writing is less than 2% after implementing of multilevel immune model. The research on biological immunity-inspired NAS devices provides a new scheme of network application for massive data storage subsystem, and makes the decentralized storage nodes have excellent performance, availability, scalability, and manageability.更多还原