Joint Scheduling Mechanism for Dynamic Slice Resource Allocation and Task Offloading in User-Edge-Cloud Systems

The heterogeneity and service diversity of vertical industries pose critical challenges for future networks, particularly in flexible resource provisioning and adaptive task scheduling. This paper proposes a hierarchical joint scheduling mechanism for inter-slice resource allocation and intra-slice task offloading in user-edge-cloud systems. A unified optimization problem is formulated by integrating slice-level and user-level scheduling slots, aiming to maximize the total long-term profit of all mobile virtual network operators (MVNOs) while reducing user delay and energy consumption. To address inter-slice scheduling, an enhanced multi-agent deep deterministic policy gradient (MADDPG) algorithm is developed, incorporating a double-target structure and prioritized experience replay to improve decision stability and convergence efficiency. For intra slice operations, a task offloading and resource allocation algorithm is developed based on deep Q-networks (DQN) and the Lagrange multiplier method. This algorithm jointly optimizes offloading ratios, offloading locations, user associations, bandwidth, and computational resource allocation. Extensive simulations demonstrate that the proposed algorithms significantly outperform benchmark schemes in terms of total profit maximization, convergence speed, and overall system performance. Furthermore, the effectiveness and scalability of the hierarchical scheduling mechanism are validated under dynamic user-edge-cloud scenarios with varying data sizes, system scales, and service demands.