Adaptive genetic algorithm for energy-efficient task scheduling on asymmetric multiprocessor system-on-chip

• Published in: Microprocessors and microsystems Vol. 66; pp. 19 - 30
• Main Authors: Yun, Yonghee, Hwang, Eun Ju, Kim, Young Hwan
• Format: Journal Article
• Language: English
• Published: Kidlington Elsevier B.V 01.04.2019; Elsevier BV
• Subjects: Adaptive algorithms; Adaptive systems; Asymmetric MPSoC; Biological evolution; Completion time; Design space exploration; Design-time task scheduling; Energy consumption; Energy optimization; Genetic algorithm; Genetic algorithms; Identification methods; Multiprocessing; Production scheduling; Static task scheduling; System on chip; Task scheduling; Design-time task scheduling; Asymmetric MPSoC; Static task scheduling; Genetic algorithm; Design space exploration; Energy optimization
• ISSN: 0141-9331; 1872-9436
• DOI: 10.1016/j.micpro.2019.01.011

•Energy-efficient task scheduling algorithm for asymmetric MPSoC, based on a genetic algorithm, is presented.•Proposed method adaptively uses three strategies during the optimization process to improve solution quality.•Proposed algorithm finds high quality solutions that reduce energy consumption of an asymmetric MPSoC while satisfying timing constraints. This paper proposes a genetic algorithm (GA) based energy-efficient design-time task scheduling algorithm, AGATS, for an asymmetric multiprocessor system-on-chip. Unlike existing GA-based task scheduling algorithms, AGATS adaptively applies different generation strategies to solution candidates based on their completion time and energy consumption. For solution candidates to evolve intelligently, instead of using conventional genetic operators, AGATS uses three generation strategies: elitism, mutation of elites (MOE), and adaptive generation (AG). The first copies a small portion of elite solution candidates into the next generation to guarantee that solution quality does not decrease from the current to the next generation. The second mutates randomly selected elite solution candidates to maintain both the diversity of candidates and solution quality. Finally, the third adaptively evolves solution candidates toward better candidates based on their completion time and energy consumption. In experiments, AGATS reduced energy consumption by up to 29.3% compared to existing methods and outperformed them in most cases. Furthermore, it identified feasible solutions effectively, which was not the case with the existing methods under tight timing constraints.