Performance analysis of software run-time behaviour using 3D-visualisation: Master's thesis

The performance tuning process of complex computing systems requires a good understanding of the run-time behaviour of the system. The increasing size and complexity of modern software highlights the importance of performance tuning as part of product development. Present day software leads to excessive energy consumption and heat production and the use of ever more powerful processors. Heat and energy issues are particularly problematic in small, mobile devices. Dynamic performance scaling technologies have been developed to limit such problems but dynamic scaling, in its turn, also increases complexity and emphasises the need for performance tuning. In this study, a visualisation and analysis tool was developed in order to enable a better understanding of run-time behaviour. A novel performance metric was also developed for evaluating and comparing different performance scaling methods independent of the execution environment, based on execution trace or theoretical analysis. The visualisation tool produces an easy-to-understand view of the trace collected from a device or simulator during execution. It shows for example the processes, processor load, performance scaling functionality and energy consumption using 3D graphics. The tool also produces a numerical output from user-selected segments of trace, including several relevant performance metrics and statistics. The applicability of the tool is demonstrated by viewing traces from real device and by analysing a simulated execution of a system containing a specific performance scaling mechanism. The effects of parameters of the scaling mechanism to the performance of the simulated system are accordingly determined and analysed