基于数据关联的分布式对象代理数据库划分方法  

Data correlation-based partition approach for distributed deputy database

作　　者：王敏[1] 彭承晨 李蓉蓉[1] 彭煜玮[1] 

机构地区：[1]武汉大学计算机学院,武汉430072

出　　处：《华东师范大学学报（自然科学版）》2016年第5期45-55,共11页Journal of East China Normal University(Natural Science)

基　　金：国家自然科学基金重点项目(61232002)

摘　　要：对象代理数据库是一种先进的具有复杂信息管理能力的数据库系统,随着数据量的剧增,实现其分布式存储变得十分重要.然而,对象代理数据库中的数据存在着很强的关联性,如果按照传统数据划分方式进行分布式存储,将导致查询效率低下.针对这一问题,本文提出了一种基于关联的高效数据划分方法:首先根据代理层次将关联对象聚集成对象簇,每个簇对应一个存储文件;然后提取对象簇的模式特征和语义特征,通过聚类算法将对象簇集划分为k个子集分配到各存储节点.将本文方法与随机分布式存储方法进行了比较实验,结果证明本文方法在查询效率方面具有明显优势.Object deputy database （ODDB） is an advanced database system with strong ability of complex information processing. With the rapid development of data, distributed storage becomes more and more important to ODDB. However, there exist correlations between objects in ODDB, which makes the traditional data portioning method of distributed storage unsuitable. To solve this problem, we propose a data correlation-based partition approach for ODDB. Firstly, we cluster correlated objects according to the deputy tree, and each object cluster is considered as a heap file in storage. Secondly, on the basis of schema feature and semantic feature, we divide object clusters into k subsets using k-means, each subset is stored on one of the storage nodes. Finally, we compare our method with random distributed storage, the results show that our approach is obviously better in query efficiency.

关 键 词：分布式 对象代理数据库 关联 数据划分 对象簇 

分 类 号：TP311[自动化与计算机技术—计算机软件与理论]