Layered Queuing Network Modeling Of Database Systems Buffer Pools

As the development and maturity of information technology, database systems have gradually become the key of information systems. In commercial areas, the amount of data is huge and complex, and fast-paced transaction processing puts forward a new requirement on the performance of database systems. The continuous development and improvement of database systems indeed meet the needs of practical applications to a large extent. However, it also increases the complexity of the systems, resulting in a system with hundreds of configuration parameters. Managing database systems has become complex and cumbersome. Relying only on database administrator to adjust and optimize database systems not only consumes financial and human resources, but also hard maintains database systems in a healthy state all the time.Artificial management of database systems limits the potential of database system, and wastes hardware and software resources. In this context, the self-managing technology of database systems came into being, also known as autonomic computing. The purpose of autonomic computing is to achieve a system with the abilities of self-configuring, self-adjustment and self-optimizing, which can control the execution of the workloads, and optimize the performance of database systems while meet the performance objectives of workloads. The core function of the technology is to achieve self-tuning and self-management of the workloads. This process needs to predict the performance of database systems in order to adjust resources allocation scheme for workloads. Building the performance prediction model of database systems is necessary.This thesis adopts the method of layered queuing network model (layered queuing networks model, referred to as LQNM), to build a model based on the allocation of buffer pool resources. Buffer pool is a region of memory allocated for the database, which is an adjustable, and has a great influence on the performance of database systems. LQNM has a wide application on the performance prediction, such as performance analysis of distributed database systems. With complex software architectures, the performance analysis of DBMS is more complex, and LQNM is applicable to the performance analysis of such complex systems. LQNM is the extension of queuing network models (queuing networks model, referred to as QNM), which can be used to analyze the competition for resources among interdependent tasks, and also can model the software and hardware at the same time, discovering the performance bottlenecks of software or hardware in the systems. But QNM can only describe the hardware resources, with the lack of support of describing of the software resources.In this thesis, the TPC-H benchmark test data is used, the experimental environment is built on the DB2. Based on the execution process of database workload and the LQNM modeling and other theories, the LQNM of database workload is built. The model parameters are obtained through a combination of the performance monitoring and the experiments, MVA algorithm is used to solve and validate the model. In the end, the impact on the performance of database systems by buffer pool resources is evaluated.