Ohjelmistoarkkitehtuuritason suorituskyvyn analysointi: Diplomityö

This work gives on overview of performance analysis at the software architectural level and methods available for that purpose. Architectural level analysis means analysing the quality of the software in the early development phase, when the system is not yet implemented and therefore the quality can not be ensured by testing or measuring the system. Instead, the analysis is conducted based on software architectural designs and estimated timing information. Software architecture means the structure of the software system, which consists of software components and interaction between these components. The basic idea related to performance analysis methods is to derive a performance model based on software architecture, which can be analysed and feedback about the performance of the planned software is obtained. Performance means the responsiveness of the software system, the ability to respond to an event or events processed in a time unit. The goal of this work was to analyse the performance of the part of a mobile phone software that is executed on the Symbian operating system during concurrent streaming and multimedia message reception. The mentioned problem area was analysed by applying the PASA method, which proposes the QNM modelling approach to be used for performance modelling, but in this work the LQN approach was used instead, because it was considered to be more suitable for modelling concurrent scenarios and layered systems. The analysis was conducted mainly by calculating utilisation, residence time and queue length based on the LQN model and estimated execution times. The calculated values were compared to performance objectives and then it was concluded that if the execution times were as estimated in this work then performance objectives would be met on average, but in the worst-case condition performance problems could occur. It was proposed in this work to change the priorities of the related tasks according to RMA principles. Then the deadlines would be met even in the worst-case. In this work performance was analysed only from the processor point of view, so the effect that other hardware resources (such as memory or buses) have on performance was not considered.