| Popis: |
A wide range of embedded systems falls into the category of safety-critical systems. Such systems impose different levels of safety requirements depending on how critical the functions assigned to the system are and on how humans interact with the system. Safety requirements involve timing constraints, the violation of which may lead to a system failure. Timing constraints are graded from soft to hard real-time constraints. While satisfying soft real-time constraints requires only best-efforts guarantees, hard real-time constraints are best-treated with worst-case analysis methods for verifying all timing constraints. Weakly-hard real-time systems have extra demands on the timing verification as they tolerate few deadline-misses in certain distributions. Applying worst-case analysis methods, in which a task is schedulable only when it can meet its deadline in the worst-case, to weakly-hard real-time systems questions the expressiveness of the computed guarantees. Considering tolerable deadline-misses raises the need for weakly-hard schedulability analyses to verify weakly-hard real-time constraints and to provide more expressive guarantees. This thesis addresses the schedulability analysis problem of weakly-hard realtime systems. It presents an efficient analysis to compute weakly-hard real-time guarantees in the form of a deadline miss model for various system models. The first contribution is a deadline miss model for a temporarily overloaded uniprocessor system with independent tasks under the Fixed Priority Preemptive and NonPreemptive scheduling policy (FPP & FPNP) using Typical Worst-Case Analysis. In our application context, the transient overload is due to sporadic tasks, for example, interrupt service routines. We adopt the proposed analysis to compute deadline miss models for independent tasks under the Earliest Deadline First (EDF) and Weighted Round-Robin (WRR) scheduling policies. In the second contribution, we extend the analysis to compute deadline miss models for task chains. The extension is motivated by an industrial case study. The third contribution of this thesis targets the system extensibility to budget under-specified tasks in a weakly-hard real-time system. Adding recovery or reconfiguration tasks such that the system still meets its weakly-hard timing constraints is of interest of an industrial case study (satellite on-board software) that is considered in this thesis. We show formally and in experiments with synthetic as well as industrial test cases that the analysis presented in this thesis can consider various scheduling policies (FPP, FPNP, EDF, WRR), and can be extended to cover both independent and dependent tasks. The thesis provides two practical solutions for two industrial case studies, which are involved exclusively in a collaboration project between Thales Research & Technology and iTUBS, which is a technology transfer company associated with Technische Universität Braunschweig. The results are thus of real practical value to be considered in the design process of weakly-hard real-time systems. |
