Genetic algorithm for DWCET evaluation on complex platform

Industrial approach to evaluate the WCET of a software test object is to benchmark it several times bare metal until enough confidence. But complexity of modern hardware and use of real-time operating system to execute multiple tasks on the top of the hardware increase the difficulty to obtain strong WCET guarantee. We propose GACC (Genetic Algorithm with Complex Contexts), a framework to evaluate the WCET of a test object using genetic algorithm but extended with the context information outside of the test software object, that is including kernel, other tasks and hardware states. As presented, this approach requires the need of coherent contexts without modeling them. Based on the need for fast evaluation, GACC records and replays hardware and software contexts, hence constructing a library of valid contexts which are used during execution of GA to find interesting solution candidates.