Adaptive Management Of Transaction Workload For Database Systems

Workload adaptation is a process in which a database management system makes use of its resources effectively to meet the performance objectives of workload by filtering and controlling workload requests. Autonomic Workload Management Framework (AWMF) is a general framework for workload adaptation, which lays out the basic components and the important processes for workload adaptation. It has been verified that the AWMF is effective for both terminal workload and batch workload. The purpose of this study is to verify that the AWMF is also effective for transaction workload. A cost-based queuing network model for transaction workload is developed and used as performance prediction component for Query Scheduler, a prototype implementation of AWMF. A set of simulation experiments is conducted to verify the effectiveness of Query Scheduler for transaction workload.By combining queuing theory and AWMF, the project established a cost-based queuing network model for transaction workload and implemented Query Scheduler. Workload control is applied at regular interval. Query Scheduler automatically allocates resources according to the performance of workload. The premise for computing workload performance is to understand the characteristics and disciplines of the workload. The population of transaction workload in a system is constantly changing, in other words, the interval of two workload requests is random. This random feature of transaction workload is described by negative exponential distribution, which is convenient to computing the workload performance.The system for experiment consists of a simulation part and real implementation part. The simulation part is used to simulate the workload and its execution, database systems, and the connection between Query Scheduler and database systems. For the transaction workload, the simulation part simulates its random feature, including the distribution of arrival interval, cost, and execute time. The real implementation part is Query Scheduler, which control the execution of workload. There are three experiments are conducted, namely, no-control, priority control, and Query Scheduler control. The no-control and the priority control experiments are the contrast for the Query Scheduler control experiment. Experimental results show that the Query Scheduler is effective for transaction workload.