Abstract:As multiprocessor real-time systems are increasingly applied in safety critical systems, it's important to ensure the correctness of such systems. One key attribute of the correctness of real-time system is the schedulability that guarantees to satisfy the timing requirements. The traditional methods for determining the schedulability are either pessimistic or unsafe. To tackle the drawbacks of those methods, this paper proposes a scheme to achieve the schedulability analysis using model checking. It provides a schedulability analysis framework for multiprocessor real-time system. All the components involved in the schedulability of the system, including the tasks, the execution infrastructure and the dispatching management unit, are modeled as timed automata and implemented in UPPAAL. Further, UPPAAL is employed to verify whether the schedulability property is always satisfied. Symbolic model checking is applied to determine schedulability. However, because of the over-approximation for stopwatches, symbolic model checking cannot be used to disprove schedulability. As a supplement, statistical model checking is used to estimate the probability of non-schedulability and generate concrete counterexamples going along with non-schedulability. Statistical model checking is also used to obtain some performance information when system is schedulable.