Towards a Graceful Degradable Multicore-System by Hierarchical Handling of Hard Errors

We present a novel concept for handling permanent faults in a statically scheduled heterogeneous multi-core system by means of a software-based self-reconfiguration. Hard faults are handled in a hierarchical cross layer manner, either locally by each core itself, or globally by reconfiguring the full system. Local reconfiguration of a defect core is based on the adaptation of the executed task to the current fault state of the core, such that defect components are never used. This adaptation is achieved by a rescheduling of the program code of the task. If this local reconfiguration fails, then the binding of tasks to cores is modified. Because heterogeneous cores are allowed, this may require a rescheduling of the tasks whose binding is changed. Estimations for the runtime of such a global reconfiguration are presented. Moreover, it is shown that systems that support the global reconfiguration achieve the same fault tolerance level as systems with local repair only, but with reduced hardware overhead.