基于Hadoop的云存储系统客户端的设计与实现-计算机系统结构专业论文.docxVIP

华中科技大学硕士学位论文 华 中 科 技 大 学 硕 士 学 位 论 文 II II Abstract Hadoop is a cloud computing platform which is the open source implementation of Googles GFS and MapReduce. Although hadoop now is mainly focused on the technology of processing of mass data calculate, we can not ignore the powerful data storage capacity of hadoop which is reflected in the design and implementation of hadoop. The powerful data storage capacity of hadoop is mainly Reflected in the design of distributed file system HDFS. HDFS is a network-based distributed file system which can handle very large amounts of data effectively, the user’ file data could be stored distributed through HDFS. In the research of hadoop platform, we mainly focus on the following three aspects. Firstly we analyze the mechanisms of data processing method and reading and writing process in the HDFS. We also analyze the defects of this design idea. The analysis will help us to build a hadoop-based cloud storage service system. Secondly we analyze the job scheduling algorithms in the hadoop platform and implement a new scheduling algorithm - based on feedback scheduling algorithm, which the main idea is the current job scheduling decisions will affect the next scheduling decision-making, and constantly modify the configuration parameters of the cluster system to achieve a more stable job processing performance. Finally, we implement a cloud storage system based on hadoop. The main achievement is the client management interface, through this platform user can manage their own storage space and file, and upload the local file to cluster servers for storage. The data of user will be processed by the HDFS before stored to the disks . Key words: cloud computing, cloud storage, job scheduling, client of cloud storage system 独创性声明 本人声明所呈交的学位论文是我个人在导师指导下进行的研究工作及取得的研 究成果。尽我所知，除文中已经标明引用的内容外，本论文不包含任何其他个人或 集体已经发表或撰写过的研究成果。对本文的研究做出贡献的个人和集体，均已在 文中以明确方式标明。本人完全意识到，本声明的法律结果由本人承担。 学位论文作者签名： 日期： 年 月 日 学位论文版权使用授权书 本学位论文作者完全了解学校有关保留、使用学位论文的规定，即：学校有权