Hard Real-Time in a Soft World

In cost conscious industries, such as automotive, it is imperative for designers to adhere to policies that reduce system resources to the extent feasible, even for safety-critical sub-systems. However, the overall reliability requirement must be both analysable and met. Faults may be either, hardware, software or timing faults. The latter being handled by hard-real time schedulability analysis, which is used to prove that no timing violations will occur. However, from a reliability and cost perspective there is a trade-off between timing guarantees, the level of hardware and software faults, and the per-unit cost. By allowing occasional deadline misses, less costly hardware may be used, while still meeting the overall reliability requirement. Careful analysis is however needed. The main risk/problem is that this type of reasoning is highly dependent on assumptions concerning distributions and independence.This paper presents a reliability analysis method that considers the effects of faults and timing parameter distributions on schedulability analysis, and its impact on the reliability estimation of the system. In scheduling terms, we will consider a wider set of scenarios/cases than just the worst case considered in hard real-time schedulability analysis. The ideas have general applicability, but the method has been developed with modelling of external interference of automotive CAN buses in mind. We illustrate the method by showing that a CAN-bus interconnected distributed system, subjected to external interference, may be proven to satisfy its timing requirements with a sufficiently high probability, even in cases when the worst-case analysis has deemed it unschedulable.