Rigorous methods for design and implementation of safety critical real-time systems are vital to avoid loss of human lives and/or severe economic losses. Unfortunately, many of these systems are designed and evaluated using ad-hoc techniques. There are, on the other hand, relatively well developed scientific theories for modeling and analysis of timing and reliability. These theories are, however, only very slowly being introduced in industrial development. Important reasons for this are the simplifying model assumptions and lack of appropriate tools for timing analysis.
This thesis presents two new methods aimed to narrow the gap between scientific results and industrial practice in evaluation and design of real-time systems.
The first contribution is a method that from execution time measurements on the target system can derive worst-case execution time estimates of programs. Such estimates are essential when verifying if a system fulfills its timing requirements. The second contribution is a simulation based technique that can be used to evaluate timing aspects of distributed real-time systems, as well as calculating reliability estimates of these systems. Such estimates are essential in determining if a system meets its requirements sufficiently well. Compared to proposed analytical methods for execution time analysis and schedulability analysis, the starting point for both these methods are real target systems, rather than an abstract model with limited correspondance to reality.
The presented initial case-studies give clear evidence that the proposed methods have potential of being both applicable and useful.
Note: Also published as report MRTC 00/25 at Mälardalens högskola
Available as compressed Postscript (375 kB)
Download BibTeX entry.