| In the field of performance evaluation of database system in the small or medium-sized enterprises, existing database benchmarks have many problems, such as meaningless metric results in practical applications, costly test implement. They also cannot discover the system bottlenecks. We present a effective way to create the domain-specific benchmarks. By this means, we solve some key technical difficulties in creating database benchmark: taking out benchmark transaction workload, describing transaction and optimizing system performance in the process of implement.We take a example of financial securities to discuss how to utilize this method in details. The dissertation mainly completes the following tasks: The paper introduces a benchmark, which is usable to measure database system performance of small or medium-sized enterprise. We define its data structure, transaction and performance metric in details. The SDB successfully solve the problems of performance evaluation in securities and other trades, facilitating to choose and improve system.Current benchmark workloads mainly generate from experience, which lacks of quantitative instruction. The dissertation quantitatively defines concepts in workload development, more accurately illustrates requirement of workload and gives the evaluation standard of workload. The paper also describes some good arithmetic to produce benchmark workload.According to theory of software engineering, combining with characteristics of workload requirement of database benchmark, the paper presents a practicable way to create workload. In this way, we integrate user behavior graphs and object-operated classifications to generate highly representative workload. In terms of this method, we get triumphantly a Securities Database Benchmark(SDB) workload.Due to amphibology of natural language, the paper defines a formal Transaction Specification Language(TSL) in order to prevent the benchmark specials. I utilize it to define more accurately workload...
|
PDF Full Text Request