Virtualized Network Function Forwarding Graph Placing with Graph Attention Network and Reinforcement Learning

Abstract

An effective method for virtualized network function forwarding graph (VNF-FG) placing based on graph attention network (GAT) and reinforcement learning (RL) is proposed for complex services and dynamic network conditions. We formulate a VNF-FG placing optimization problem, and designed a GAT-based RL agent to recognize graph structure and obtain placing policy. Experiments prove the method effectiveness.