基于对象存储文件系统研究

发布时间：2018-06-01 17:34

本文选题：基于对象存储 + 文件系统　；参考：《电子科技大学》2012年硕士论文

【摘要】：现今，，越来越多行业加快了信息化建设，导致数据量成倍增加；另一方面，电子制造技术的不断突破发展，使得CPU速度已不再是计算机性能的瓶颈，存储介质价格越来越便宜、容量越来越大。此外，网络技术的飞速发展更是让数据量成指数增长，存储空间不断扩充，网络存储面临巨大的挑战。但是目前的网络存储技术使用的是传统接口，导致无法有效应对不断增长的存储需求。基于对象的接口，因为同时具备文件和块接口的特性，成为应对存储压力的有效方法，受到学术界和企业界广泛而深入的研究。基于对象存储系统（OBS）通过基于对象的接口，将文件系统的存储管理功能下移到对象存储设备（OSD）中，元数据服务器只需简单的将用户对文件的请求转化为对应的对象标示符，用户通过对象标示符直接与OSD设备进行数据交互，从而有效减轻了元数据服务器的负载，提高了元数据服务器的响应速度以及实际数据的传输速度。业界对基于对象存储的研究主要集中在元数据服务器的设计和整体性能的提升上，对单个OSD设备上的本地文件系统研究很少。通常是在OSD设备上直接使用传统文件系统结合对象逻辑层的方法来进行存储管理，这样无法充分发挥基于对象接口的优势，制约了OSD设备的性能，同时，当OSD设备不断增加时，单个OSD设备性能的损失也会严重影响整体性能。为了解决上述问题，本文以基于对象的本地文件系统ZFS为对象，分析对象存储的实现思想，并加入文件类型识别功能，使得ZFS文件系统能根据文件类型进行智能的存取，从而提高单个OSD设备上的存取性能。论文主要工作和成果如下： 1.学习基于对象的理论，分析OSD协议、文件类型识别方法； 2.阅读ZFS源码，提取ZFS文件系统中关键数据结构，分析主要操作的实现过程，理清ZFS代码整体流程； 3.提取文件类型特征标识符，设计实现文件类型识别模块，并加入到ZFS代码中来实现ZFS智能存取； 4.通过实验验证本文对ZFS文件系统所做改进。
[Abstract]:Nowadays, more and more industries speed up the construction of information technology, resulting in the increase of data volume. On the other hand, with the development of electronic manufacturing technology, the speed of CPU is no longer the bottleneck of computer performance, and the price of storage media is becoming cheaper and cheaper. The capacity is getting bigger and bigger. In addition, with the rapid development of network technology, the amount of data increases exponentially, the storage space expands continuously, and the network storage is faced with enormous challenges. However, the current network storage technology uses traditional interfaces, which can not effectively cope with the growing demand for storage. Object-based interfaces, because of the characteristics of both file and block interfaces, have become an effective method to deal with storage pressure, which has been extensively and deeply studied by academia and business circles. OBS-based storage system (OBS) moves the storage management function of the file system down to the object storage device (OSDs) through the object-based interface. The metadata server only needs to simply convert the user's request to the file into the corresponding object identifier. The user interacts with the OSD device directly through the object identifier, which effectively reduces the load of the metadata server and improves the response speed of the metadata server and the transmission speed of the actual data. The research of object based storage is mainly focused on the design of metadata server and the improvement of overall performance, but little on the local file system on a single OSD device. Usually, the traditional file system combined with object logic layer is used directly on the OSD device for storage management, which can not give full play to the advantage of the object based interface, which restricts the performance of the OSD device. At the same time, when the OSD device continues to increase, The performance loss of a single OSD device can also seriously affect overall performance. In order to solve the above problems, this paper takes the object based local file system (ZFS) as the object, analyzes the realization idea of the object storage, and adds the file type recognition function to make the ZFS file system access intelligently according to the file type. This improves access performance on a single OSD device. The main work and results are as follows: 1. Learn object based theory, analyze OSD protocol, file type recognition method; 2. Read the ZFS source code, extract the key data structure in the ZFS file system, analyze the implementation process of the main operations, and clarify the overall process of ZFS code; 3. Extract the file type identifier, design and implement the file type identification module, and add it to the ZFS code to realize the intelligent access of ZFS; 4. The improvement of ZFS file system is verified by experiments.
【学位授予单位】：电子科技大学
【学位级别】：硕士
【学位授予年份】：2012
【分类号】：TP333

【参考文献】