边缘环境下硬件感知的任务卸载方法OA

Existing studies on Mobile Edge Computing(MEC)often assume hardware-agnostic task workloads,overlooking the effects of thread-level parallelism and CPU micro-architectural factors on execu-tion.This simplification leads to errors in task completion-time prediction,ultimately affecting the accuracy of offloading decisions.To address this issue,this paper proposes a hardware-aware task offloading method(HAODQ).First,we propose a hardware-aware multi-threaded parallel computing model by considering CPU micro-architectural parameters such as pipeline depth and cache hit rate,and incorporating task thread-level parallelism and multicore resource contention constraints.This model can capture both the computational demands of tasks and the computing power of computing nodes more accurately,enabling precise estimation of task execution time.Second,based on the hardware-aware computing model,we formulate a task offloading model for concurrent task execution in MEC systems by integrating communication and system models with the objective of minimizing the task completion time.Third,we develop a deep reinforcement learning(DRL)-based task offloading algorithm that leverages a Deep Q-Network(DQN)to match tasks with edge servers intelligently.Experimental results demonstrate that the proposed algorithm significantly reduces the task completion time and improves the overall performance of the MEC system compared with other algorithms.