Research On The Autonomic Database Systems: Theories And Approaches

As Internet and information technology develops, database systems are getting more and more complicated, and data scale in the database system is getting larger and larger. It results in increasing cost and complexity to manage a database system. Autonomic database systems can automatically and actively manage themselves to reduce the complexity of database management. The employment of autonomic database systems can also decrease the manpower cost and extra cost brought by system failure. Therefor, autonomic database techniques are vital for database systems in critical applications, and these techniques can vastly reduce the total cost of ownership (TCO) and lessen the waste of hardware and software investment. This thesis mainly focuses on the theories, techniques and implementations of autonomic database system.First, related areas of autonomic computing and autonomic database systems are thoroughly investigated. Current research issues, methodologies and commercial applications in these areas are analyzed. Major problems of current techniques are identified, which arouse following research in this thesis.Based on analysis and pattern extraction of the artificial database management process by DBAs, concepts related with database management are defined. The manageable autonomic database element (MADE), an universal meta-structure with six components and a knowledge base, is introduced to support solving database self-managing problems. In this thesis, MADE is used to study three kinds of autonomic features in database system: automatic physical database design, adaptive database security management, and adaptive load balancing in parallel database system.Automatic physical database design techniques for integratively designing b+trees, vertical partitions and attribute compression are studied. Based on the formalization of the automatic physical database design problem, the PD-MADE, one kind of MADE, is proposed to solve the problem. Four key techniques in PD-MADE are mainly investigated:workload analysis based on SQL classification and transaction clustering, rule-based physical feature generation, cost estimation for data volume and workload processing, and optimal configuration search based on greedy algorithms and swarm evolution algorithms.For adaptive database security management, concepts of dynamic threats, static threats and security policies are defined. It is deemed that security policies are two-faced, i.e. each security policy will settle some threats, but simultaneously increase the probability and severity of some other threats. Based on these concepts, the adaptive security management problem is defined, and SM-MADE, another MADE, is proposed to solve the problem. Intrusion detection, security policy selection and security policies are covered, and database intrusion detection techniques based on extraction of transaction features and categorization of transaction vectors is studied in detail.An autonomic parallel database architecture is proposed, and two kinds of load balancing problems, the data migration problem and the parallel join problem, are introduced. The LB-MADE, a MADE for solving the load balancing problems, is studied, and mainly concentrates on the node selection and execution techniques of data migration and parallel join process.Based on above studies in this thesis, we implement three kinds of experiment systems: the OSCAR physical database advisor, an OSCAR database intrusion detection prototype, and a prototype of autonomic parallel database middleware. Experiments' results on these systems validate the autonomic database architecture, the manageable autonomic database element, and three kinds of specified autonomic techniques proposed in this thesis.