A Reinforcement Learning Approach for Multipath TCP Data Scheduling
Jiacheng Luo, Xin Su, Bei Liu · 2019
Multi-path TCP(MPTCP) is a new multipath transmission control protocol which can improve the throughput by aggregating resources of the different paths together. However, the performance of MPTCP is seriously affected by the path environment. With the bounded resources and heterogeneous network, the default Round-Robin data scheduling(RRDS) algorithm in MPTCP transmits data packets to every path in turn will degrade the throughput of MPTCP in many situations. In order to solve this problem, we propose a new data scheduling algorithm based on reinforcement learning(RL) with the newly introduced Deep Q Network (DQN) framework to enhance the MPTCP data scheduling performance in the asymmetric path. The reinforcement learning algorithm gets the information of every path and adaptively choose the most suitable path by the artificial intelligence. The simulation results show that the proposed data scheduling algorithm with RL can significantly achieve a higher throughput than the default algorithm.