基于Chord的多租户索引机制研究  

Indexing Mechanism of Multi-Tenant Database Based on Chord

作　　者：邹立达[1,2] 李庆忠[1,2] 孔兰菊[1,2] 王振坤[1,2] 

机构地区：[1]山东大学计算机科学与技术学院,济南250101 [2]山东省软件工程重点实验室,济南250101

出　　处：《计算机学报》2016年第2期270-285,共16页Chinese Journal of Computers

基　　金：国家自然科学基金(61572295;61303085);山东省自然科学基金(ZR2013FQ014;ZR2014FM031);山东省科技发展计划(2014GGX101047);山东省自主创新专项项目(2015ZDJQ01002;2015ZDXX0201B03)资助~~

摘　　要：多租户数据管理是支持SaaS应用高效运行的重要组成部分.随着租户规模的不断扩大,多租户数据库需要云计算环境下的每个节点都存储并处理租户数据,因此适合利用对等结构(P2P)组织管理多租户数据.组织良好、易扩展的云中多租户索引机制是高效查询的关键.文中基于P2P结构对多租户索引机制展开研究,针对不同租户的索引易相互干扰、租户数据分布无序的问题,通过对Chord的映射方法进行改进将所有租户索引统一映射到一个标识符空间,给出的映射函数使单个租户索引可以隔离、保序地在空间分布.同时设计了各节点所负责的标识符空间范围的分配算法,使租户索引及数据可均衡、聚集地放置在各节点,从而使查询时既能通过对等结构避免性能瓶颈,也降低了数据传输成本.文中给出了多租户索引机制的动态维护策略,提出的标识符空间增倍方法使该索引机制能够适应租户数量与索引不断增加的应用场景.实验结果表明,该机制在租户规模较大时,与集中式索引相比查询时间至少可以节省50%,吞吐量提高1.5倍.Multi-tenant data management is an important part of supporting SaaS applications torunefficiently. With the continuous expansion of the tenants, multi-tenant database is deployed in cloud computing environment. Since every node in clouds stores and processes multi-tenant data, it needs a peer to peer （PZP） structure to organize multi-tenant data management system. In this case, a well-organized, scalable multi-tenant indexing mechanism in clouds is critical for efficient querying. In this paper we proposed a multi-tenant indexing mechanism supporting P2P structure. For solving the problems of mutual interference of multi-tenant indexes and disordered distribution of multi-tenant data, we improved the mapping approach of identifier space in Chord, which makes a single tenant index be stored in the cloud, in isolation and in order. Moreover, we designed an assignment algorithm of identifier space to evenly and clustered place multi-tenant indexes and data to nodes, which not only avoids bottleneck of querying performance through P2P structure, but also reduces data transmission cost. Last we presented a dynamic maintenance strategy of multi-tenant indexing mechanism to adapt to the scenario of increasing indexes and tenants by doubling identifier space. Through the extensive experiments, we demonstrate that the proposed mechanism outperforms centralized indexing mechanism with at least 50% of query time reduction and 1.5 times throughput increasement.

关 键 词：多租户索引 P2P CHORD 多租户数据库 标识符空间 云计算 

分 类 号：TP393[自动化与计算机技术—计算机应用技术]