A Study Of Traffic Allocation Mechanisms For Multi-path Networks

With the rapid development of the Internet and various new technologies,more and more user devices support multiple ways to access the Internet,but the traditional TCP(Transmission Control Protocol,TCP)protocol can only establish one connection at a time for communication,which greatly reduces the efficiency of communication,and multipath transmission technology has gradually attracted the attention of researchers.As a result,researchers have proposed Multipath TCP(MPTCP),a protocol based on TCP,which can establish multiple communication links for data transmission and improve the overall utilization of the network.The traditional MPTCP cannot handle different kinds of data flows in the network well,resulting in some delay-sensitive flows not completing transmission in a timely manner and increasing the completion time of data.At the same time,in actual multipath networks,paths are often asymmetrically heterogeneous,i.e.,there are relatively large differences in delay,bandwidth and packet loss rates between paths,leading to poor data scheduling during data transmission by MPTCP,resulting in packet disorder at the receiving end and even blocking in the buffer.Therefore,this thesis will provide different transmission strategies in asymmetric heterogeneous networks according to the characteristics of different kinds of flows and the round-trip delay and congestion level of sub-flows.The existing data scheduling algorithm is also improved to reduce the packet disorder and solve the problem of blocking in the buffer.Based on the above analysis,the main research of this thesis is as follows:(1)To address the problem that MPTCP uses all available paths to transmit different kinds of data streams,resulting in increased completion times for delay-sensitive streams.In this thesis,a transmission optimization strategy based on data stream classification is proposed.First,a quantitative analysis is performed based on the round-trip delay and congestion level of each path,and an optimal path set is established by selecting the paths with small round-trip delay and small congestion level.Then on the basis of the optimal path set,a suitable transmission strategy is selected for the data streams.For short streams that are delay sensitive,the best sub-streams in the path set are dynamically selected for transmission using Markov decision to minimize completion time;for long streams that value throughput,the optimal path set is used for transmission to ensure the throughput of long streams,so that different kinds of streams can be well transmitted and transmission efficiency is optimized.(2)To address the problem that MPTCP transmits data on asymmetric paths,resulting in packets reaching the buffer at the receiving end out of order and causing buffer blocking.In this thesis,a dynamic packet scheduling algorithm is proposed,which first combines the round-trip delay of the path and the congestion window to reasonably allocate the packets sent for each path by predicting the number of packets that can be sent by fast and slow sub-streams,so that the packets arrive at the receiving end in order.Then,on this basis,the scheduling value is adjusted by feedback based on whether the SACK selection confirmation signal is received,so that the scheduling value of packets is more accurate,thus reducing the impact on future rounds.The problem of buffer blocking caused by packet disorder is effectively solved.Finally,the proposed algorithm is tested and analyzed by simulation software.The proposed algorithm is found to be better than the default MPTCP algorithm in handling different types of data streams and solving packet disorder.The proposed algorithm can reduce the completion time of short streams by about 30% and effectively reduce the number of disordered packets,which has better transmission performance.