基于HBase的数据压缩技术研究

  • 格式:pdf
  • 大小:1.67 MB
  • 文档页数:71

研究生签名:____________ 导师签名:____________ 日期:_____________
万方数据
摘要
随着大数据技术的发展以及 Hadoop 等大数据平台的迅速普及与推广, 生活中产生的数据 量呈现爆炸性增长的趋势,数据种类呈现复杂化,存储方式呈现多样化。传统的基于行存储 的大数据存储方式并不能够以较低的成本将大数据存储起来。与此同时,由于数据的访问频 度的不同,对于不同访问级别的数据所采用的存储方式提出了新的要求。针对以上情况,结 合大数据平台下的 HBase 数据库,本文对大规模数据环境下基于 HBase 的压缩存储技术进行 了研究,主要的创新点如下: 首先,提出一种基于访问频度的数据分类方法:根据一段时间内数据库文件的访问次数 得到相应的访问频度,依据各数据文件的访问频度及相关阈值将数据文件划分为冷热数据并 确定具体的访问级别。在此基础之上,提出基于数据访问级别的压缩策略选择方法:定义了 确定数据样本的抽样方法,针对原有的压缩策略选择方法中先验知识未必可靠的缺陷,通过 添加评估层及时调整先验知识,并在基于相邻参照区和基于统计列选择方法的基础上设计出 HBase 数据压缩策略选择方法,优化存储成本。仿真实验与结果表明,本文提出的方法不仅 能够有效实现大数据的存储,同时还提高了数据的访问性能。 其次,从数据迁移的角度,提出一种基于文件价值的数据迁移方法。首先,根据数据访 问频度等因素计算出数据块文件的价值,由这个文件价值得到数据迁移的目的设备。同时改 进了数据迁移技术,利用数据缓冲区和双缓冲队列解决了数据迁入迁出速率不匹配的问题, 提高了数据迁移效率,节省了内存和时间消耗,最终实现了对大数据平台数据的存储优化。 最后,基于以上的方法与理论,本文构建了基于数据压缩存储的原型系统并给出一个电 子商务应用示范。系统的实现遵循需求分析、概要设计、详细设计及其实现等流程,完成压 缩存储管理、数据迁移等功能模块,验证了本文提出算法的可行性,展现了基于 HBase 的压 缩技术理论成果在动态场景下的应用效果。
By Caihang Fu Supervisor: Prof. Haiyan Wang March 2016
万方数据
南京邮电大学学位论文原创性声明
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,除了文中特别加以标注和致谢的地方外,论文中不包含其他人已经发表或撰写过 的研究成果,也不包含为获得南京邮电大学或其它教育机构的学位或证书而使用过的材料。 与我一同工作的同志对本研究所做的任何贡献均已在论文中作了明确的说明并表示了谢意。 本人学位论文及涉及相关资料若有不实,愿意承担一切相关的法律责任。
II
万方数据
process, such as complete compression storage management, data migration, such as function modules, the feasibility of the proposed algorithm is verified, the results showed the compression technology based on HBase theory in dynamic situations of application effect. Key words:cold and hot data, data access level, HBase,data coon
单位代码: 10293

级:
硕 士 学 位 论 文
论文题目: 基于 HBase 的数据压缩技术研究
学 姓 导 学 研 科 究 专 方
号 名 师 业 向
1013041118 伏彩航 王海艳 软件工程
_ _ _ _
网络环境下软件理论与技术 _ 工学硕士 2016 年 3 月 _ _
申请学位类别 论文提交日期
关键词: 冷热数据,访问级别,HBase,数据压缩,数据迁移
I
万方数据
Abstract
With the development of the technology of big data and the rapid popularization and promotion of Hadoop platform, in daily life the amount of data shows the tendency of explosive growth, many types of data are becoming complicated and storage ways of data are various. Traditional way of large data storage is not able to lower the cost of large data stored. At the same time, because of the different data access frequency, the way of data storage of different access level are different. In view of the above situation, this thesis based on the compression of HBase under the environment of mass data storage technology are studied, the main innovation points are as follows: First of all, the thesis puts forward a data classification method based on the access frequency: according to the number of visits the database file over a period of time to get the corresponding access frequency, according to the access frequency of each data file and related threshold could be divided into hot and cold data file data and determine the specific level of access. On this basis, put forward based on the data access level compression strategy selection method: defines the data samples to determine the sampling method, in view of the original compression strategy choice method of defects of the prior knowledge is not necessarily reliable by adding evaluation layer adjust prior knowledge, and based on the reference region and the adjacent column selection method based on the statistics on the basis of design the HBase data compression strategy selection methods, optimize the storage costs. Simulation experiments and the results show that the proposed method not only can effectively realize large data storage, but also improve the performance of data access. Secondly, from the perspective of data migration, the thesis puts forward a method based on the value of the file data migration. First of all, based on factors such as data access frequency calculated the value of a block of data files, and by the value of the file to get the purpose of data migration. At the same time improve the data migration technology, using the data buffer and double buffer queue to solve the data into emigration rate mismatch problem, improve the efficiency of data migration, saving memory and time consumption, finally achieved the big data platform data storage optimization. Finally, based on the above theory and method, this thesis built a prototype system based on data compression storage and an e-commerce application demonstration. The realization of the system follows the requirements analysis, general design, detailed design and its implementation