Mixed Criticality Systems: Beyond Transient Faults

Adopting mixed-criticality architectures enable safe sharing of computational resources between tasks of different criticalities consequently leading to reduced Size, Weight and Power (SWaP) requirements. A majority of the research in mixed-criticality systems focuses on scheduling tasks whose Worst Case Execution Times (WCETs) are certified to varying levels of assurances. If any given task overruns its WCET, the system switches to a higher criticality and all the lower criticality tasks are discarded to make time for the execution of higher criticality tasks. Task execution time overruns are transient faults that are typically tolerated by simply executing an alternate task before the original deadline, or, by discarding the failed task to prevent it from interfering with higher criticality tasks. However, permanent faults such as processor failures can render the system to be useless, many times leading to unsafe states. In this paper we present a taxonomy of fault tolerance techniques to tolerate permanent faults, as well as map it to real-time mixed-criticality requirements based on the extend of fault coverage that in turn influences the associated assurance.