Abstract :

    In this thesis, we investigate several approaches aimed at validating the temporal constraints of real-time systems. This work is applied to in-vehicle embedded applications distributed on a CAN network. The validation process is performed using different and complementary techniques: simulation, analysis and prototype monitoring. The main contribution is the design of analytical models that provide bounds on the considered performance metrics (response time, deadline failure probability) or that enable us to evaluate the occurrence of rare events (bus-off hitting time of a CAN node). We also propose a schedulability analysis of real-time applications running on Posix1003.1b compliant operating systems. Considering that usually there exist several solutions to the same scheduling problem, we define general criteria for chosing among feasible solutions as well as a genetic algorithm for exploring the solution space. The second objective is to address the problem of jointly scheduling hard real-time and soft real-time traffic with different performance objectives: ensuring that the timing requirements of hard real-time traffic are met while minimizing as much as possible the response time of soft real-time traffic. We first evaluate the performance of the Dual-Priority policy for the scheduling of messages. Then we suggest a simple mechanism that provides probabilistic guarantees to prevent hard real-time frames from missing their deadlines under unreliable transmission. We also provide an on-line adaptive procedure for setting the parameters of the error model which is well suited for systems where the bus perturbation level may vary greatly over time. Finally, we propose a traffic shaping policy having the same goals as Dual-Priority but more easily applicable. This policy preserves feasibility and its complexity being linear in the number of hard real-time messages, it can be used on-line.

Keywords : Validation, Optimisation, Real-Time, Scheduling, Embedded Systems.