A dynamic reliability management framework for heterogeneous multicore systems

Dynamic Reliability Management (DRM) is an attractive system-level approach to mitigate the effects of aging and degradation phenomena in systems equipped with multiple computing resources, trading performance to improve lifetime reliability and/or power efficiency. In this paper, we propose a novel DRM controller for Heterogeneous System Architectures running on the top of Linux Operating System able to dynamically adapt to the system conditions and workload characteristics, leveraging on the applications' performance, power consumption and system lifetime. Based on the user's goal, the framework can be exploited to implement DRM policies optimizing the desired trade-off between the identified metrics. The implementation of the controller on the Samsung Exynos 5 architecture is reported together with the evaluation of a state-of-the-art mapping policy here enhanced to be reliability-aware.