| Real-time systems must react continuously to their environments within time constraints. The use of such systems is rapidly growing; these systems are widely applied in various applications such as consumer electronics, industrial automation, and medical instrumentation. The design objectives are strongly influenced by the requirements of their non-functional constraints, such as timing and memory constraints. For example, high-integrity applications, where failure could cause loss of life or environmental harm, have high development and maintenance costs. The success of real-time embedded systems relies mostly upon their capability to produce functionally correct results within defined timing constraints. Therefore, timing analysis is crucial to be able to guarantee that all hard real-time tasks meet their deadlines in accordance with the design requirements.;In this dissertation, we present an analytical-based approach to estimating the worst-case response time (WCRT) for a set of real-time tasks having fixed or dynamic-priorities. The motivation for this work comes primarily from the study of the abort-and-restart (ANR) scheduling model, which is neither a concurrency control policy nor a true scheduling policy. Instead, it is a policy in which the most important tasks are scheduled first. Our research addresses the specific issue of computing an upper bound on the worst-case response time of a task considering the direct interference of higher-priority tasks and the abort and restarts that they cause. An important performance aspect of the abort-and-restart scheme is that the model reduces the priority-inversion problem, but makes reasoning about the overall system correctness more difficult. We show that in some cases the number of aborts is limited by the event set attributes or characteristics, so the handlers' response times are bounded. In other cases the number of aborts is high. There is more overhead with the abort-and-restart model, since time spent on task execution is wasted. However, the model does reduce the impact of a wrong decision. It implements executions as function evaluations. Therefore, there are no "incomplete" functions evaluations as in the case of a task's interruption. We argue that the added computational costs is reasonable, since the model offers far richer composition than traditional scheduling schemes. |
