Database architecture design and performance evaluation for Quality of Service measurements: Master's thesis

Measuring and analysing the quality of experience and the quality of service are important assisting methods when optimizing network capacity, and providing the best possible user experience. They also present methods for finding and dealing with problems caused by the ever-growing internet data traffic and its number of users. Qosmet, developed at VTT Technical Research Centre of Finland Ltd, is a powerful solution for this type of network analysis. Considering that long term storage of quality of service measurement data is all numerical, this enables an efficient and well optimised database solution, one that performs well in either writing data or reading it for analysis. The purpose of this thesis is to ascertain the optimal database architecture for quality of service measurements. An important outcome of this project is the interface between Qosmet and the implemented database. Part of the work consists also in identifying the characteristics for database optimization. In order to achieve these goals, three database architecture schemas were designed and analysed with the help of exhaustive writing and reading performance tests. Schema 1 has one, schema 2 two and schema 3 have respectively three database tables, in which the data is distributed according to its assumed frequency of use. Following the analysis and comparison of these three database architecture schemas, the one with two tables appeared to be the best one. It achieved writing throughput of around 7,500 lines per second, and the reading performance was the second best, actually very close to the best performance. Schemas 1 and 2 were about 33 percent faster than schema 3 in writing performance. Schemas 2 and 3 performed about 20 percent better than schema 1 in reading tests. Performance analysis gave a good understanding of database optimization and features affecting its efficiency, as well as hardware requirements, bottlenecks and limits on high utilization situations. This work resulted in an efficient and verified database, which can be used in the future as part of the Qosmet solution.