Research Of Data Scheduling And Congestion Control Algorithms On Intelligent Mptcp

With the increasing demand for high-speed transmission and network bandwidth in industries such as webcasting and drones,traditional single-link transmission protocols are unable to achieve the required uplink and downlink network speeds.Moreover,modern devices have multiple network interfaces such as 4G,5G,and Wi-Fi,but only one interface can be used at a time,which wastes network resources.To address these issues,researchers have proposed Multi-Path TCP(MPTCP)protocol that utilizes multiple network interfaces of a device to establish multiple transmission paths between the client and the server.This approach allows the full utilization of network bandwidth resources,leading to an increased throughput of data links and meeting users’ demands for high-bandwidth networks.The multipath nature of MPTCP protocol enables the use of multiple links for transmission,enhancing the robustness of the transmission links.As such,there has been much research in the area of multipath transmission.To ensure link fairness and solve the link congestion problem and the packet disorder problem in MPTCP,this paper proposes algorithm designs for both congestion control and data scheduling processes.The primary goal is to achieve real-time multipath data transmission.(1)An Adaptive dynamics Bandwidth Estimation based on Coupling of Asynchronous Advantage Actor-Critic scheduling algorithm(ABEA3C)is proposed in the congestion control aspect.Firstly,the bandwidth estimation algorithm based on link state accurately estimates the current bandwidth value of the link and rapidly relieves the link congestion and restores the link to normal use.Subsequently,the asynchronous feature of the A3 C algorithm is leveraged to train multiple intelligences for accurately determining the congestion state of the link.Based on the estimated bandwidth,the transmission rate of the link is assigned accordingly.Finally,using the gradient update algorithm to train ensure that they make the correct decision every time,thereby improving the throughput as well as the robustness of the link.(2)A Path Dynamics Assessment Asynchronous Advantage Actor-Critic scheduling algorithm(PDAA3C)based on A3 C is proposed in the data scheduling phase.Then the packets are allocated according to the path state,employs multi-threaded training ideas from asynchronous algorithms to allocate packets based on the path state and the transmission attribute value of the link.This approach improves the accuracy of the link state evaluation and makes packet allocation more reasonable and precise in the data scheduling stage.Thus making the packet allocation in the data scheduling stage more reasonable and accurate,and thus improving the link throughput and fairness of the link.Simulation results show that,compared with the MPTCP-BBR and MPTCP-Vegas algorithms,the ABEA3 C algorithm for network link bandwidth estimation can steadily converge the bandwidth estimate to the ideal bandwidth value and increase the throughput of the network link by about 30%,thus effectively alleviating the network congestion;the PDAA3 C algorithm performs link state evaluation before packet distribution This improves packet arrival disorder and ensures link fairness.Throughput is improved by 26.6% ～ 106%compared to MPTCP-Round-Robin and MPTCP-Fastest-RTT,and by 8.6% compared to the MPTCP-RLDS algorithm combined with deep learning,and fairness is close to the theoretical optimum 1.