On the limits of probabilistic timing analysis
- Autores: Suzana Milutinovic
- Directores de la Tesis: Francisco J. Cazorla (dir. tes.), Jaime Abella Ferrer (codir. tes.)
- Lectura: En la Universitat Politècnica de Catalunya (UPC) ( España ) en 2019
- Idioma: español
- Tribunal Calificador de la Tesis: Luca Santinelli (presid.), Miquel Moretó Planas (secret.), Carles Hernández Luz (voc.)
- Programa de doctorado: Programa de Doctorado en Arquitectura de Computadores por la Universidad Politécnica de Catalunya
- Materias:
- Texto completo no disponible (Saber más ...)
- Resumen
Over the last years, we are witnessing the steady and rapid growth of Critica! Real-Time Embedded Systems (CRTES) industries, such as automotive and aerospace. Many of the increasingly-complex CRTES' functionalities that are currently implemented with mechanical means are moving towards to an electromechanical implementation controlled by critica! software. This trend results in a two-fold consequence. First, the size and complexity of critical-software increases in every new embedded product. And second, high-performance hardware features like caches are more frequently used in real-time processors.
The increase in complexity of CRTES challenges the validation and verification process, a necessary step to certify that the system is safe for deployment. Timing validation and verification includes the computation of the Worst-Case Execution Time (WCET) estimates, which need to be trustworthy and tight. Traditional timing analysis are challenged by the use of complex hardware/software, resulting in low-quality WCET estimates, which tend to add significant pessimism to guarantee estimates' trustworthiness. This calls for new solutions that help tightening WCET estimates in a safe manner.
In this Thesis, we investigate the novel Measurement-Based Probabilistic Timing Analysis (MBPTA), which in its original version already shows potential to deliver trustworthy and tight WCETs for tasks running on complex systems. First, we propose a methodology to assess and ensure that ali cache memory layouts, which can significantly impact WCET, have been adequately factored in the WCET estimation process. Second, we provide a solution to achieve simultaneously cache representativeness and full path coverage. This solution provides evidence proving that WCET estimates obtained are valid for ali program execution paths regardless of how code and data are laid out in the cache. Lastly, we analyse and expose the main misconceptions and pitfalls that can prevent a sound application of WCET analysis based on extreme value theory, which is used as part of MBPTA.
