Research On The Fairness Mechanism Of MPTCP

With the development of communication technologies and the falling price of elec-tronic devices,communication terminals usually have multiple interfaces(such as 4G and WIFI).There are usually multiple communication paths between two communica-tion terminals.As a new transport layer protocol,MPTCP has the ability to coordinate the use of multiple paths.Compared with traditional TCP,it has higher throughput,better reliability,and it can move data packets from crowded paths to uncrowded paths.To ensure the fairness and friendliness towards TCP,one of the design goals of MPTCP is to satisfy the network fairness criterion,that is,the throughput of a MPTCP connection should be equal to the throughput of a TCP connection on the best paths of MPTCP.Existing congestion control algorithms fail to satisfy this requirement during Slow-Start phase.During initial Slow-Start phase,a MPTCP connection can achieve more throughput than TCP.Unfairness becomes even severe as the number of subflows increase.In addition,the serious packet losses during Slow-Start often leads to poor performance of MPTCP.We find that MPTCP is even worser than TCP.In order to address the above issues,we present the Coupled Slow-Start(CSS)Algorithm for MPTCP,which couples subflows of MPTCP during the initial Slow-Start phase.CSS links exponential growth of subflows' congestion windows and resets Slow-Start Threshold(ssthresh)of different subflows to give the MPTCP connection as much throughput as TCP,aiming to ensure fairness and reduce packet losses.It can improve performance of MPTCP during Slow-Start.In addition,although the network fairness criterion guarantees the fairness of MPTCP towards TCP,it limits the performance of MPTCP when its subflows have distinctive bottlenecks.In such cases,there is no need to ensure fairness towards TCP,but MPTCP still has to meet the criterion of total throughput being no more than TCP,which limits the performance of MPTCP.Therefore,recent studies have proposed a bottleneck fair-ness criterion:the sum of throughput of subflows which share a common bottleneck should be equal to the throughput that TCP can achieve through that bottleneck.For the remaining subflows there is no such restriction.Each of these subflows can achieve as much throughput as TCP.This makes MPTCP's shared bottleneck detection a very critical issue.To improve accuracy of shared bottleneck detection of MPTCP and accelerate the implementation and deployment of bottleneck fairness based congestion control algo-rithm.We present a variance-based shared bottleneck detection scheme(VSBD).VSBD computes the standard deviation of time interval between the two flows experiencing congestion events at the receiver and compares it with the squared mean of sum of the duration of corresponding congestion events.If the standard deviation is less than squared mean,it is determined that the two flows share a bottleneck;otherwise it is de-termined that the two flows have distinctive bottlenecks.Simulations show that VSBD has high detection accuracy in various scenarios,especially when the bottlenecks are partially overlapped.