With the rapid growth of the Internet of Things, computation‑intensive directed acyclic graph (DAG) tasks are becoming common in cloud‑edge‑end collaborative settings. The heterogeneous computing power, network bandwidth, and energy consumption of cloud, edge, and end nodes turn the scheduling of dependency‑rich tasks into an NP‑hard problem. Conventional heuristics and standard reinforcement‑learning approaches often fail to capture the spatio‑temporal dynamics of system resources.
This work introduces PPO‑STGNN, a DAG scheduling algorithm that fuses proximal policy optimization (PPO) with spatio‑temporal graph neural networks (STGNN). The STGNN extracts features from both the DAG task topology and the physical resource graph, after which PPO optimizes the scheduling policy to minimize makespan and schedule length ratio (SLR) while improving CPU and memory load balancing. A multi‑teacher behavior‑cloning scheme is employed for pretraining to accelerate convergence. Experiments demonstrate that PPO‑STGNN markedly enhances load balancing while keeping completion time low, making it suitable for dynamic and heterogeneous cloud‑edge‑end DAG scheduling scenarios.
Review