ATMP: An Adaptive Tolerance-based Mixed-criticality Protocol for Multi-core Systems

The challenge of mixed-criticality scheduling is to keep tasks of higher criticality running in case of resource shortages caused by faults. Traditionally, mixed-criticality scheduling has focused on methods to handle faults where tasks overrun their optimistic worst-case execution time (WCET) estimate. In this paper we present the Adaptive Tolerance-based Mixed-criticality Protocol (ATMP), which generalises the concept of mixed-criticality scheduling to handle also faults of other nature, like failure of cores in a multi-core system. ATMP is an adaptation method triggered by resource shortage at runtime. The first step of ATMP is to re-partition the task to the available cores and the second step is to optimise the utility at each core using the tolerance-based real-time computing model (TRTCM). The evaluation shows that the utility optimisation of ATMP can achieve a smoother degradation of service compared to just abandoning tasks.