Energy-aware task mapping combining DVFS and task duplication for multicore networked systems

Integrating high-performance communication and computation capabilities, multicore embedded platforms have become key components to realize applications of networked systems, e.g., Cyber-Physical Systems (CPS). Such systems usually consist of multiple dependent and real-time tasks that can be executed in parallel on different cores of the nodes and have timing, energy, and reliability constraints. Designing efficient task mapping methods to transmit and process task data under multiple constraints is challenging. Existing works seldom consider the joint design problem under timing, energy, and reliability constraints, which are coupled with each other, introducing complexity in designing efficient task mapping methods. In this paper, we first formulate the joint design problem as a complex combinational optimization problem and design a linearization method to find the optimal solution. To reduce computation complexity and enhance scalability, we design a decomposition-based heuristic method. Then, a refinement method based on feedback control is added to enhance task schedulability. The results show that the optimal solution obtained by the proposed method achieves the desired system performance. Moreover, the proposed heuristic provides a feasible solution with negligible computing time (reduces 99.9% computation time but with 24.3% performance loss). Compared with the existing works, our method can optimize the usage of system resources to balance energy, timing, and reliability requirements.