Research On A Multi-Hierarchical Data Distribution Model

Today, relational databases faces an unprecedented challenge in massively distributed sys-tems like cloud computing, due to the difficulties of data distribution, data consistency, access control, and scalability etc. Key/value databases have been widely used in such systems. In this dissertation, a more abstract data distribution model(Multi-Hierarchical data Distribution model) is builded on top of the relational model to solve these difficulties, resulting in following findings.Local closed word assumption being introduced in this paper, a large scale distributed data system can be viewed as a huge complex system composing many independent sub-systems, in which data nodes are controlled by several data nodes, implementing data organizing and distribution.A multi-hierarchical data distribution model is defined in this paper. The method of re-stricting data graph be data multitree is presents through a theorem, and for schema graph with circuits or diamonds due to sematic restrictions, methods of approximating data graph to data multitree are investigated. The referential integrity of this data distribution model is presented too, including ancestral integrity and control integrity. The operations on data multitree is also touched on, including union, intersection, difference, merging, narrowing, extraction and base-line.Based on the Multi-hierarchical data distribution model, a data consistency model(fuzzy timestamp multi-version) is proposed, as well as corresponding data replications and update on data replicas. On the analysis of the restrictions of traditional serializablity theory, a transaction model fitting local closed word assumption is presented. The relation between the data con-sistency model and the transaction model, and the isolation levels are also investigated to meet transaction consistency.A access control model based on the separation between operations and data is also pro-posed, which is consistent with the multi-hierarchical data distribution model. User's region is defined, region, user, role,and operations being investigated, defining virtual users to simplify the security management of large scale distributed systems.The multi-hieratical data distribution model can improve the scalability of relational database. After the scalability being quantitatively defined, a general method of improving scal-ability is given in the dissertation. MHM can conveniently implementing scale horizontally and vertically. Based on TPC-C benchmark database, and Mysql cluster platform, compare exper-iments are done on 4,8,16,24,32 nodes respectively. The results show that MHM is apparently more scalable than relational model, since MHM can prevent system saturation earlier.In summary, MHM is compatible with the relational model, improve scalability, and is also a security integrated data distribution model. MHM can meet the need of large-scale distributed systems.