Tolerating sensor timing faults in highly responsive hard real-time systems

Real-time systems that have to respond to environmental state changes within a very short latency period often use event-triggered task activation. If the system has to function correctly in the presence of sensor faults, event-triggered task activation is not reliable. Faulty sensors may cause task activations to occur too early, too late, or task activations are omitted entirely. In particular, early task activations can overload the system. Time-triggered task activation is reliable, but by defining a competitiveness ratio it is shown that the processor utilization for highly responsive tasks is unacceptably low. To overcome the problems of event-triggered task activation while preserving its good performance the task-splitting model is introduced. The task-splitting model integrates fault tolerance into the analysis and construction of hard real-time systems by using a combination of event-triggered and time-triggered task activation. Based on a general task model, it is independent of any particular scheduling algorithm. The result of this work has influenced the design of a new operating system which will be applied in a robust automotive engine controller of the next generation