Safe Multi-Agent Deep Reinforcement Learning for Dynamic Virtual Network Allocation

Network traffic and computing demand have been changing dramatically due to the growth of various types of network services, e.g., high-quality video delivery and OS update. To maximize the utilization efficiency of limited network resources, network resource control technology is required for smooth and quick operation when the network demands change. We propose a dynamic virtual network allocation method based on safe multi-agent deep reinforcement learning (safe MA-DRL). This method can quickly optimize network resources even while network demands are drastically changing by learning the relationship between network demand patterns and optimal allocation by using the DRL algorithm in advance. We developed two techniques to be used with our method; safety-considerations and multi-agent. Our safety-considerations technique reduces the degree of constraint violations, such as network congestion and server overload, and our multi-agent technique improves the scalability of virtual network allocation by dividing demands into groups and assigning each group's allocation to each agent. As a result of a simulation evaluation, safe MA-DRL can calculate effective allocation within 1 s that doubles the link utilization efficiency without any constraint violations compared to the static virtual network allocation method.