Energy Optimization and Fault Tolerance to Embedded System Based on Adaptive Heterogeneous Multi-Core Hardware Architecture
Xing Liu, Panwen Liu, Xinping Yan, Chengming Zou, Ruoshi Xia, Haiying Zhou, Kun Mean Hou, Christophe de Vaulx · 2018
Energy optimization and fault recovery are significant to ensure the embedded devices to work persistently and reliably. To achieve the energy optimization, a reconfigurable heterogeneous multi-core hardware platform is designed. Based on this hardware platform, a multi-core energy-efficient scheduling mechanism using reinforcement learning algorithm to search for the optimal scheduling solution is proposed. By the above hardware and software collaborative optimization mechanism, the energy cost of embedded system can be decreased significantly. To perform the fault recovery, a multi-core control-flow fault recovery mechanism is researched. This mechanism uses the Petri net model to detect the multi-core control-flow faults, and then recover these faults by a hardware-based quick recovery technique. The experimental results showed the energy cost of EMWSN could be optimized by more than 20% comparing to the traditional multi-core system. In addition, about 90% multi-core control-flow faults could be recovered, and the recovery time was nearly 40% less than the software-based recovery technique CFCSS.