A Deep Graph Reinforcement Learning Approach for Multi-streaming Scheduling in Real-Time Communication Network

Real-time communication network have put forward higher requirements for latency and stability of data transmission. Multi-streaming transmission mechanism can concurrently transmit data on a single connection, so it’s widely used in various protocols in real-time communication network. However, the congestion and retransmission of a single stream will still cause the aggregated throughput to decrease. In this paper, We propose a multi-streaming scheduling method based on deep graph reinforcement learning (DGRL), which outputs optimized multi-streaming transmission rate adjustment strategy by real-time acquisition of runtime state information, and improves the aggregated throughput of data transmission in real-time communication network. The simulation results shows that the proposed algorithm can effectively improve the throughput under network fluctuation environment.