|
Abstract:
|
Currently surprisingly few publications exist on the subject of practical performance measurement. Much of the existing material focuses on narrow and specific applications or on hardly comprehensible mathematical formulas. As the complexity of embedded systems increases and speed requirements of applications get constantly tougher, the need for easy to use methods for performance verification becomes ever more important. In this thesis, the concept of the response time is discussed, and to help analyse it two more key concepts are discussed: CPU utilization (the time-load) and memory usage (the memory-load). The relation between these concepts is then studied to some depth based mostly on the RAM theorems. Measurement methods are then developed with the selected performance factors. Simplicity and ease of use are kept in mind throughout the development process, and so the resulting methods require minimal or no modifications to the measurement target. One key idea is to make performance measurement so simple that it can be performed along all normal testing and thus a tool for follow up.Finally, a wider scope is taken towards performance measurement. The entire process, form test requirement setting to evaluating the results is discussed. /Kir10 |