This report is a presentation of a new approach for Worst Case Execu- tion Time (WCET) computation for hard real-time systems, especially for cache memories hazard issues. A general overview of issues and state of the art in this matter is drawn, but the main point is the theory in itself and its formalism, first, in single task execution, then in multitasking environ- ment. The method being used relies on abstract interpretation, like many other methods of WCET computation, but its formalism is a probabilistic approach (although it is deterministic in the single task field) with the use of Markov chains. Generalization to the multitask field make use of this prob- abilistic base to compute a pessimistic evaluation of the WCET and of its standard deviation, thanks to a clever modification of the propagator. First evaluation of the model results, hand-coded from the compilation output of rather simple tasks, shows that results are promising in order to apply it to real-life applications.
