Research on Resource Reservation Strategy for Edge Federation

With the increasing number of users and the continuous expansion of task scales, the resource constraints faced by edge computing are becoming increasingly pronounced. To address these challenges, edge computing has gradually evolved into an edge federation computing model. This model enhances overall resource utilization through multi-node resource sharing within the federation, effectively alleviating the issue of insufficient resources at individual edge nodes. This paper first analyzes the necessity of implementing a resource reservation strategy within the edge federation. It proposes a load prediction-based resource reservation strategy and optimizes this strategy according to the load levels present in the edge federation. Subsequently, we introduce a multi-agent deep deterministic policy gradient (RRP-MADDPG) approach based on a multi-agent deep reinforcement learning algorithm aimed at reducing average task delay. Simulation results demonstrate that both proposed resource reservation strategies can significantly reduce average task delays. Furthermore, the RRP-MADDPG strategy exhibits excellent convergence performance and outperforms both the load prediction-based reservation strategy and other similar deep reinforcement learning algorithms.