A multi-layer software-based fault-tolerance approach for heterogenous multi-core systems

This paper describes a software-based technique for building heterogeneous fault tolerant multi-core systems, which are able to handle temporary and permanent hardware faults autonomously in two system layers. The fault tolerance technique relies on a single concept for adapting the binary code of the user application to the current fault state of a single core. Thereby this scheme is used either for a local repair of each core or for a global repair. By the global repair, the task assigned to a faulty core may be rescheduled to another core that provides enough resources for the execution of the task. Thereby the local repair scheme is reused for the adaptation of the rescheduled task. It is shown that the reliability of a multi-core system can be improved significantly, when using the global repair together with the local repair instead of using the local repair only.