A Reinforcement Learning-based Intelligent Communication Scheduling Framework for Enhancing Coordination in Distributed Systems

The efficiency of communication scheduling has a huge impact on the performance of large-scale distributed systems, especially under dynamic conditions like node failures and network fluctuations. Traditional scheduling mechanisms that rely on static rules are difficult to adapt to such environments. This paper proposes an intelligent communication scheduling framework based on reinforcement learning, which takes Qlearning as the core and combines with a decentralized Gossip network to achieve dynamic optimization of communication node selection. Compared to complex deep reinforcement learning algorithms, Q-learning has low time complexity, concise implementation, and low computational resource overhead, making it more suitable for communication scheduling tasks in lightweight distributed environments. The framework enables communication strategies to continuously adaptively adjust during training and respond to network changes in real time through system state modelling and reward feedback mechanisms. The experiments are carried out in an environment that simulates multiple topologies with different node numbers and different failure rates. The results show that the method significantly outperforms the traditional Gossip protocol in terms of communication overhead and number of messages. However, it still fluctuates in terms of convergence error, especially in high failure rate scenarios. The study further analyses the limitations of Q-learning in partially observable state spaces and proposes an improvement direction based on deep reinforcement learning and reward structure optimization.