Adaptive offloading scheme of fine-grained tasks based on reinforcement learning in multi-access edge computing

In the context of multi-access edge computing, the constraints on device resources and energy pose a challenge in meeting the requirements of delay-sensitive and high-resource demand tasks. In order to address this issue, this paper puts forth an adaptive offloading decision based on fine-grained task sequencing (AOD-FTS). To enhance task execution efficiency and resource utilization, we design scheduling optimization for subtasks algorithm within the fine-grained task model to optimize the execution order of subtasks and improve subtask parallelism. Additionally, to reduce task delay and energy consumption, we propose an adaptive multi-objective reinforcement learning offloading algorithm based on ordered subtasks, formulated through multi-objective Markov decision processes. The scheme dynamically adjusts optimization preferences according to environmental changes and task completion states to make offloading decisions. Simulation results demonstrate that compared with two benchmark algorithms, the proposed AOD-FTS scheme achieves: (1) higher task completion rates, (2) improved resource utilization efficiency, (3) 81.66 % and 51.53 % reductions in delay respectively, and (4) 39.72 % and 71.36 % reductions in energy consumption respectively.