GCN-Based Multi-Agent Deep Reinforcement Learning for Dynamic Service Function Chain Deployment in IoT
Shuyi Wang, Haotong Cao, Longxiang Yang, Sahil Garg, Georges Kaddoum, Mubarak Alrashoud
Abstract
The rapid development of technologies such as the Internet of Things, SDN/NFV, and 6G is driving up the demand for dynamic deployment of service function chains (SFC). These technologies are making network architectures more complex and service deployments more dynamic and adaptable. More than ever, there are situations that call for multi-objective SFC dynamic deployment, which necessitates resource game optimization across multiple objectives. For the first time, multi-objective optimization in dynamic SFC deployment scenarios is realized using a multi-agent deep reinforcement learning system based on graph convolutional network (GCN) in this study. Here we mainly focus on the game optimization problem of two objectives: minimum delay time and minimum resource utilization. Three sample complex networks are used to evaluate the proposed methodology: Random, BA scale-free, and Small-world. The results of the simulation indicate that the proposed method can be well applied in IoT scenarios. In general, this method is superior to other mainstream methods in terms of reward and convergence performance.