Deep Deterministic Policy Gradient based Dynamic Virtual Network Embedding Algorithm
Yue Zong, Han Xu, Zhaoyang Zhang · 2023
Due to the limitation of network ossification, network virtualization is a promising architecture to solve the issue. Virtual network embedding is one of the challenges of network virtualization. However, existing algorithms cannot solve the online dynamic requests and intelligent embedding solution selection. Deep reinforcement learning technology have been used to solve the issues in communications and network area due to its good features to do decision-making. However, many of the literatures just utilize Q-learning or DQN to solve the VNE issue. In this paper, we propose a Deep Deterministic Policy Gradient (DDPG) based virtual network embedding mechanism for resource allocation and management for dynamic online virtual network requests. The proposed algorithm can improve the dynamic scheduling of virtual network embedding. Simulation results have evaluated that the proposed DDPG-VNE algorithm can significantly improve the acceptance ratio, the profit cost ratio compared with benchmarks.