System-level timing feasibility test for cyber-physical automotive systems

Abstract

For automotive systems there is a mismatch between worst-case timing analysis models and the perceived reality, diminishing their relevance, especially for the automotive powertrain domain. Strict worst-case guarantees are rarely needed in the powertrain domain. The reason is that a large amount of functionality is control software and this can tolerate sporadic deadline misses. For instance, certain control approaches can systematically account for sampling losses and still prove whether or not the controller is stable and adheres to required performance criteria. Typical worst-case analysis (TWCA) tackles this problem by providing formal guarantees on typical response-times including upper bounds on the number of violations of these. In this paper, we derive a system-level timing feasibility test exploiting the robustness of control applications based on TWCA. We extend the TWCA to cope with periodic tasks that have varying execution times. Taking the robustness of control applications into account, we derive upper bounds for the overload models of each task, along with possible typical worst-case execution times (TCET), as needed for the TWCA. We then use this information to find a feasible typical-case configuration such that all deadlines are reached and all robustness constraints are satisfied. To verify the approach and show the expressiveness, we apply it on a performance model of a full-blown modern engine management system provided by Bosch.