A Research On Congestion Control Of MPTCP

In recent years, with the development of internet technology, communication terminal equipments typically have multiple network interfaces, which can make the communication between the hosts simultaneously use more than one network card and path. The experiments can get more fast recovery capabilities, security and load-sharing capabilities when the end-hosts are multihomed. Multipath TCP, as proposed by the IETF working group in 2009. The IETF is responsible for the standard of the MPTCP, solving the problem of the congestion control and routing during the transmission and making it comparable with the traditional TCP. It is not only increasing the bandwidth of the network, improving the utilization of the resources effectively, but also enhancing the service transmission rate and load balancing capabilities by aggregating the sub-flows. In addition, the MPTCP can move the traffic from more congested links to parallel ones timely, which can effectively improve the robustness of the network and ensure the reliability of data.This paper makes a study on the methods of the multi-path parallel protocol congestion control from two aspects: fairness and balancing congestion. With respect to fairness, the paper proposed DWBC(Dynamic Window Based Coupling) algorithm. The Algorithm is based on the principle of fairness.Through this method, the congestion window can change dynamically by setting a parameter of a, aggregating sub-streams together by detecting shared bottleneck link. DWBC can dynamically intergrate the different sub-streams of the congestion window into a collection, using RTT compensation mechanism for the same combination of a sub-stream to ensure the fairness of TCP. The algorithm can adapt to changes in the network, and can quickly recover from the error. Under four simulation scenarios, DWBC can do a good performance.From equilibrium congestion, the paper propose a congestion control algorithm which is based on packet loss rate(Drop-based LIA). The congestion windows can be classified into different types through this algorithm.The method can increase and decrease the size of the congestion window through the accelerating factor and reduction factor. The network can get stable in a fast speed and get a better resourcing pool state. Also through the different scenarios simulations, DBLIA can be more efficient with the two factors than LIA.