The introduction of 5G/B5G in vehicular networks has created more and more traffic in backbone networks. As a result, the datacenter traffic optimization becomes a crucial issue in satisfying various services to mobile users. Traditionally, datacenter traffic was either configured by network administrator or by a fixed TCP congestion control protocol. Unfortunately, the manual solution is limited by the experience of the administrator and obviously not scalable. As for TCP congestion protocols, they are simply a set of simple rules, and as a consequence may not capture the complexity of network traffic at datacenter. We believe that the recent advancement in artificial intelligence, specifically in the area of Deep Reinforcement Learning, is suitable for traffic control and optimization at datacenter, as TCP acknowledgements can be utilized as the feedback in policy learning mechanism. Deep Reinforcement Learning, such as the one used for AlphaGo in DeepMind, has been shown to be both versatile and robust in recent studies. Depending on the number of the datacenters, different Deep Reinforcement Learning schemes (e.g., single-agent or multi-agent) can be properly adopted. To be specific, we propose the following works: 1) Building Datacenter Simulation Environment, i.e., the “Playground” in Reinforcement Learning; 2) Single-Agent Deep Reinforcement Learning Traffic Control for One Datacenter; 3) Multi-Agent Deep Reinforcement Learning Traffic Control for A Cluster of Datacenters; 4) Solution Verification via Testbed. These works joint together aims at developing a practical and viable Deep Reinforcement Learning solution for datacenter traffic optimization.
|Effective start/end date||1/08/21 → 31/07/22|
In 2015, UN member states agreed to 17 global Sustainable Development Goals (SDGs) to end poverty, protect the planet and ensure prosperity for all. This project contributes towards the following SDG(s):