Towards Smart Stream Scheduler for Multipath QUIC in Heterogeneous Networks

In recent years, with the rapid development of internet communication technology and the continuous increase of terminal users, the amount of data transmission carried by the internet has increased rapidly, and users’ requirements for service quality have also become higher. Multipath QUIC (MPQUIC) can utilize multiple network interfaces of terminal devices for data transmission, increasing bandwidth utilization, improving network fault tolerance and robustness. However, heuristic schedulers based on fixed parameters or models cannot adapt to dynamic heterogeneous network environments. In this paper, we proposed a Smart Stream Scheduler (SmSS) that combines with deep reinforcement learning Double Deep Q-Network (DDQN). SmSS uses DDQN to capture the relationship between environmental states and stream scheduling decisions, so as to match the most suitable stream with idle paths based on the current environmental state, thereby improving transmission performance. The experimental results based on Mininet show that in dynamic heterogeneous network environments, compared to the default scheduler LRF of MPQUIC, SmSS can reduce transmission delay by up to 8.2%.