Congestion control algorithms are critical for ensuring efficient and reliable data transmission in data center networks.However,traditional algorithms with fixed parameter settings often fail to achieve high performance in dynamic network environments.Existing reinforcement learning-based congestion control schemes,while effective,typically rely on centralized training methods,leading to data silo issues and privacy concerns associated with sharing raw data.To address these challenges,this paper proposes a federated reinforcement learning-based congestion control framework.This framework leverages the advantages of distributed network architectures by treating each network device as an independent agent and enabling collaborative training among agents through federated learning techniques.This approach not only protects the privacy of each participant's data but also enhances the adaptability and generalization of the model by integrating the knowledge and strategies of individual agents.As a case study,we apply this framework to explicit congestion notification threshold optimization,and experimental results across various scenarios demonstrate the superior performance of the proposed framework.
关键词
数据中心网络/拥塞控制/强化学习/联邦学习
Key words
data center networks/congestion control/reinforcement learning/federated learning
维普
万方数据