Modeling Database System Performance With Locking Description

With the development of information technology, optimizing application performance has become paramount important for enterprises. The optimization of database systems involves dynamic adjusting and tuning parameters and control workloads, which has a significant impact on the performance of application sand occupies the core position in the optimization of application performance. Optimizing the performance of database systems usually needs a model to guide the optimization. Therefore, establishing a precise performance model of database systems has become the key task for optimizing the performance of database systems.Locking as a part of internal of database systems, its impact on the performance of database systems cannot be ignored. At present, the main technology on locking is queuing networks model, but it only describes computer hardware resources, and cannot describe the hierarchical structures and software resources existing modern computer systems, it has influence on the analysis and description the flow of the lock.This thesis uses the layered queuing network model for lock modeling. Layered queuing network model, an extension of queuing network model, is designed to be able to model the software resources, in addition to hardware resources. Layered queuing network model can better describe and analyze the dependence of calls and hierarchical relationships among resources and tasks, making up the deficiency for queuing network model. At the same time, if the row locks are too many the database systems will run out of memory, which will result in the row locks automatically upgraded to table locks, for this account this thesis only models table locks. The main work as follows:1. By analyzing the locking mechanism, a layered queuing network model with locking description is proposed for database systems. The experiment employs TPC-H Benchmark as the data and workloads, and DB2as the target database system.2. The model is solved and analyzed using the method of layered. The model parameters obtained by using the direct and indirect methods. Due to load the concurrent access will cause the lock waiting, however, the competition for resources also cause the lock waiting in actual execution time. Therefore, the experiments use the1-0.5of the proportion of lock waiting time in total time to solving the lock waiting time.3. The model is solved and analyzed using the three sets of load. By comparing with the queuing network model the proposed layered queuing network model is more accurate.