【摘要】 针对传统的负载编排策略没有充分考虑对算力资源的编排，导致系统整体能效较低的问题，本文提出一种算力能效驱动的负载编排策略，对边缘环境中的网络和计算资源协同编排。首先，对算力环境中的计算资源进行信息熵算力度量，构建边缘负载模型和算力能效模型以提升计算资源利用率。其次，考虑到边缘环境的动态特性，将负载编排问题建模为马尔可夫决策过程，设计一种基于柔性演员评论家(SAC)的深度强化学习方法来优化能效并且提供优化的任务卸载决策。最后，实验验证所提出的算力能效驱动的负载编排策略在计算资源利用率、能耗、任务完成率等方面优于传统策略。更多还原

【Abstract】 Aiming at the problem that traditional workload orchestration strategies did not fully consider the orchestration of computing resources, resulting in low overall energy efficiency of the system, this paper proposed a computing energy-efficiency driven workload orchestration strategy to collaboratively orchestrate network and computing resources in edge environments. Firstly, the information entropy-based computing power measurement of computing resources was conducted in the computing environment, and an edge workload model and a computing energy-efficiency model were established to improve the utilization of computing resources. Secondly, considering the dynamic characteristics of edge environments, the workload orchestration problem was modeled as a Markov decision process, where a soft actor-critic(SAC)-based deep reinforcement learning method was designed to optimize energy efficiency and provide optimal task offloading decisions. Finally, experimental results demonstrate that the proposed compulitity energy-efficiency driven workload orchestration strategy outperforms traditional approaches in terms of computational resource utilization, energy consumption, and task completion rate.更多还原