Deep Reinforcement Learning-Based Age Aware Task Scheduling Algorithm for Computing Power Network

Abstract

Computing power networks (CPNs) have emerged as a promising paradigm for supporting delay-sensitive and freshness-critical IoT services in the 6G era by jointly orchestrating heterogeneous networking and computing resources. However, existing scheduling schemes mainly focus on offloading decisions and computation-node selection, while ignoring the execution-stage scheduling process after tasks arrive at the selected computation node. Such an omission may lead to queueing delay and stale task execution, which can severely impair the reliability of real-time IoT applications. To address this problem, this paper proposes a freshness-aware task scheduling scheme for CPNs, where task offloading, network path selection, and execution-stage scheduling are jointly considered. A twin delayed deep deterministic policy gradient (TD3)-based algorithm is developed to make adaptive scheduling decisions according to dynamic network states, computing-resource availability, queueing conditions, and Age of Information (AoI). Simulation results show that the proposed scheme achieves better scheduling performance than benchmark algorithms, demonstrating its effectiveness in supporting freshness-sensitive IoT applications over CPNs.