A Hybrid Multi-objective Evolutionary Algorithm for Energy-Aware Allocation and Scheduling Optimization of MPSoCs

Rongjie Yan, Yupeng Zhou, Yige Yan, Minghao Yin, Min Yu, Feifei Ma, Kai Huang

Abstract: MPSoCs are increasingly being adopted in the design of emerging complex embedded systems. Resource limitations require designers to find optimizations among various design considerations. Task mapping and scheduling become one of the key issues in designing such systems. To meet the requirements of makespan minimization and workload balance for energy-aware MPSoCs, the paper presents a unified formulation to find satisfied task mapping and scheduling solutions. The model considers both computation and communication cost, and enables applying dynamic power management (DPM) for energy optimization. To efficiently approximate the Pareto front of the optimization problem, we propose a multi-objective hybrid algorithm (MOHA) by integrating a Pareto local search into an evolutionary process, with a problem-specific initialization. Experimental results from realistic benchmarks demonstrate that the proposed techniques are able to generate high-quality solutions of realistic applications on the target architecture, compared with state-of-the-art methods.