MA-DHRL-OM: Multi-Agent Deep Hierarchical Reinforcement Learning for Overlay Multicast Routing
A novel multi-agent deep hierarchical reinforcement learning method, named MA-DHRL-OM, has been introduced for Overlay Multicast (OM) path planning. Traditional OM faces challenges with dynamic traffic due to a lack of awareness regarding physical resource conditions, while current reinforcement learning techniques struggle to address OM's interconnected multi-objective complexity, leading to increased difficulty, slower convergence rates, and instability. By utilizing the global perspective provided by SDN, MA-DHRL-OM constructs a traffic-aware model, breaking down the OM tree creation into two phases through hierarchical agents, which minimizes the action space and enhances convergence stability. Collaborative efforts among multiple agents facilitate multi-objective optimization and boost scalability and adaptability. Experimental results indicate that MA-DHRL-OM surpasses existing approaches in terms of delay and bandwidth efficiency.
Key facts
- MA-DHRL-OM is a multi-agent deep hierarchical reinforcement learning method for Overlay Multicast routing.
- It uses SDN's global view to build a traffic-aware model for OM path planning.
- The method decomposes OM tree construction into two stages via hierarchical agents.
- It reduces action space and improves convergence stability.
- Multi-agent collaboration balances multi-objective optimization.
- Experiments show MA-DHRL-OM outperforms existing methods in delay and bandwidth utilization.
- The approach addresses traditional OM's inability to adapt to dynamic traffic.
- Existing reinforcement learning methods fail to decouple OM's multi-objective nature.
Entities
—