SaaS多租户数据副本管理机制研究

发布时间：2018-03-27 17:32

本文选题：SaaS　切入点：多租户　出处：《山东大学》2014年硕士论文

【摘要】：SaaS是通过因特网向外界提供服务的软件运营模式,利用多租户技术支持灵活性租赁和按需定制,作为二十一世纪云服务平台中的全新型应用交付模式,越来越受到各企业的追捧。 SaaS平台多采用副本技术提高节点冗余度,以保障多租户数据的可靠性与可用性。当一些数据副本出现故障或失败时,剩余副本能够继续提供服务,以便更好地保护租户数据,提高系统容错性和数据可用性。引入副本技术将会产生一系列新问题：1)传统的数据副本模型，，无法针对多租户数据即共享又隔离的存储特征,设计动态副本管理模型,更不能适时定位多租户共享架构的副本单位,来适应不同租户的SLA；2)多租户数据请求消息,在不同的时间间隔内,数据请求消息到达的个数没有必然的联系,这种多租户数据请求特征无法使用传统的基于历史数据的预测机制来计算多租户数据副本数量。3)由于无法扑捉多租户数据的特征,现有的副本放置策略,很难保证副本节点间良好通讯及减少分布式事务的发生概率。随着多租户数据的运行,租户数据和租户数量在不断增加,现有的数据副本调整策略,很难适应多租户共享数据的热点分布,不能及时合理地进行数据副本的调整,解决热点瓶颈,提高数据可用性。针对以上问题,本文首先探讨了SaaS软件交付平台的多租户副本模式,提出面向租户的动态副本管理框架,并针对多租户数据特征,提出适合多租户数据的副本数量预测、放置与调整机制。本文的主要贡献如下： (1)针对多租户数据存储特征,提出一种面向多租户的动态副本管理模型,解决了多租户数据共享宽表存储模式下的副本管理问题。本文综合多租户数据副本管理所需功能,深入探讨动态数据副本管理模式,设计出依附于SaaS软件交付平台的多租户数据副本动态管理框架,并给出框架中各组件之间关系以及各组件在多租户数据请求消息访问过程中所起的作用。本文还改进传统分布式网络拓扑结构,提出分层网络模型,解决了单点瓶颈,环形热点问题。 (2)针对多租户数据请求消息分布现状,提出基于消息队列的租户数据副本预测技术,解决了无法完全依赖历史数据预测租户数据副本数量的问题,提高了租户数据副本管理的准确性和高效性。本文首先分析多租户数据节点在云平台环境中的故障率,在保证多租户数据可用性的前提下,计算出平台运行初期的最低副本数量阈值。接着深入研究多租户数据请求消息的分布情况,得出请求消息的概率分布。然后采用面向多租户数据请求消息的消息队列模型,计算出平台所需的接近最优的副本数量。 (3)在面向多租户的动态副本管理框架的基础上,提出多租户数据副本放置与调整算法,解决了因不合理数据放置与调整引起的租户数据倾斜、节点间通讯阻塞、分布式事务增加等问题。本文根据多租户数据存储结构的共享特征,提取影响租户数据放置与调整的因子,统筹计算出适合租户数据副本放置的节点位置。并进一步通过消息过滤,消息队列优化与轮转,消息响应时间监听方式进行三阶段数据副本调整,以保证及时并适时地进行租户副本调整,降低出现租户数据可用性不能满足的概率。本文提出依附于SaaS软件交付平台的多租户数据副本管理框架,通过预测数据副本数量,统筹放置数据副本,及时调整副本数据,降低不满足租户可用性要求、租户分布式事务、额外副本调整发生的概率,减少节点间通讯费用,节省系统资源,平衡系统负载,提高租户数据可用性。
[Abstract]:SaaS is a software operation mode that provides services through the Internet to the outside world. It uses multi tenant technology to support flexible leasing and on-demand customization, and is a new mode of application delivery in the cloud service platform in twenty-first Century. It is attracting more and more attention from all enterprises.
SaaS platform uses copy technology to improve the node redundancy, in order to ensure the reliability and usability of multi tenant data. When some data replica failure or failure, the remaining copies can continue to provide services, in order to better protect the tenants' data, improve the system fault tolerance and data availability.
The introduction of replica technology will produce a series of new problems: 1) a copy of the data model of the traditional storage characteristics to multi tenant data sharing and isolation, the design of dynamic replica management model, not timely positioning multi tenant sharing architecture a copy of the unit, to adapt to different tenants SLA; 2) multi tenant data request message and in different time intervals, not necessarily related to the number of data request message arrives, the multi tenant data request features are not available in traditional prediction mechanism based on historical data to calculate the multi tenant data copy number.3) due to the inability to capture characteristics of multi tenant data placement strategy, the existing copy is very difficult ensure good communication between the replica node and the probability of occurrence of the reduced distributed transaction. With multi tenant data operation, the number of tenants and tenants in the data increases, a copy of the data transfer current The whole strategy is difficult to adapt to the hot distribution of multi tenant sharing data. It can not adjust the data copy in time and reasonably, solve the hot spot bottleneck and improve the availability of data.
To solve the above problems, this paper first discusses the SaaS software delivery platform of multi tenant copy mode, put forward dynamic replica management framework for tenants, and for the multi tenant data characteristics, put forward the suitable number of copies of the multi tenant data prediction, placement and adjustment mechanism. The main contributions are as follows:
(1) for multi tenant data storage characteristics, a multi tenant dynamic replica management model is proposed, which solves the problem of replica management in multi tenant data sharing and wide table storage mode.
The multi tenant data replica management functions for the in-depth study of dynamic data replication management mode, designed the attachment delivery platform in SaaS software multi tenant data copy dynamic management framework, and gives the framework of relationship between various components and components in the rent according to the number of request messages in the process of visiting the role. The improvement of the traditional distributed network topology, the layered network model, solves the single point bottleneck, the annular hot issue.
(2) in view of the current situation of multi tenant data request message distribution, a message queue based tenant data replication prediction technology is proposed, which solves the problem that the number of replica data can not be completely dependent on historical data, and improves the accuracy and efficiency of tenant data replica management.
This paper first analyzes the multi tenant data node failures in the cloud platform in the environment of multi tenant rate, under the premise of ensuring the availability of data, calculate the minimum number of copies of the platform running initial threshold. Then we study the distribution of multi tenant data request message, the probability distribution of the request message. Then use the multi tenant data request message queue model, calculate the number of copies of the near optimal platform is required.
(3) on the basis of multi tenant dynamic replica management framework, a multi tenant data replica placement and adjustment algorithm is proposed, which solves the problem of tenant data skew due to unreasonable data placement and adjustment, communication blocking between nodes, and distributed transaction increase.
According to the features of sharing multi tenant data storage structure, extracting data placement and adjustment factor tenants, and calculate the node location for tenant data replicas. And further through message filtering, message queue optimization and rotation, the response time of news supervision adjusted data copy of the three stage, to ensure timely and timely for a copy of the tenant adjustment, reduces the probability of the occurrence of the availability of data can not meet the tenants.
This attachment multi tenant data replica management framework delivery platform in SaaS software, the prediction data copy number, and placed a copy of the data, timely adjust the copy of the data, reduce the availability requirements to meet the tenants, tenants distributed transactions, the probability of occurrence of the extra copy adjustment, reduce communication costs, save system resources, balance system load, improve the tenant data availability.

【学位授予单位】：山东大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TP393.09

【参考文献】