Importance-Based Scheduling to Manage Multiple Core Defection in Real-Time Systems.

This paper presents an approach to support multiple permanent node failures in multicore real time systems. In the absence of failures, the system is scheduled with the PFair algorithm PD2. To overcome failures, a single spare core is provided and two protocols based on task importance are defined: The Recovery Time Distribution Protocol (RTDP) and the Graceful Degradation Protocol (GDP). When a single core fails, RTDP sets the system parameters such that all the tasks still meet their deadlines, although after a bounded delay. When several cores fail, GDP defines several modes corresponding to degraded execution. Different strategies are provided to decide which tasks are dropped in degraded modes. The experimentation of both protocols shows conclusive results. Tasks recover from the failure in a bounded delay with RTDP, whereas there are some missed deadlines with GDP. However, we exploit the experimental results to guide the designer on which elimination strategy to use.