Research On Coding For End-to-end Transmission Protocols

TCPis the most important transport protocol today and the majority of data in the Internet are transmitted through TCP currently. With the rapid development of network technologies and devices, modern networks such as the mobile Internet, high speed data center networks and heterogeneous multi-path networks are more and more popular, whose characteristics and application requirements are far different from those when TCP was initially designed. TCP has seen performance limitations on throughput and delay in these networks. Its multipath extension version MPTCP is then proposed to provide concurrent multipath transmission, improving the goodput and reliability of TCP by making full use of network resources, but introduced new challenges on the buffer size requirement.This thesis focuses on alleviating the limitations of standard TCP and its multipath extension version by applying coding, and designs a coding-based end-to-end transport protocol with dynamic coding schemes as well as a coding-based multipath protocol with an intelligent data allocation algorithm.We firstly focus on the usage of coding in regular single-path TCP to improve the performance through providing redundant transmission and forward error correction. To make better use of coding, we model the packet loss and coding in TCP, and then design a dynamic coding scheme by scheduling the block size and coding redundancy to adapt to the varying network conditions. Analysis and simulation results show that through intelligent scheduling, coding can maximize the bandwidth utilizationwhile providing an appropriate trade-off between the goodput and delivery delay.Then we propose to tackle the performance degradation problem of multipath TCP in heterogeneous multipath networks by taking advantage of rateless coding. We conduct an extensive simulation-based study on that problem, and the results show that a low-quality subflow with higher delay and loss may severely affect the other subfows especially when the buffer size is limited, thus becoming the bottleneck of the whole multipath connection.We then design a Fountain code-based Multipath TCP denoted FMTCP to mitigate the negative impact of the heterogeneity of different subflows.Moreover, we estimate the expected arriving time of subflows by combining the round trip time, loss rate and window size, and design an algorithm to perform data allocation among subflows based on the estimation. Quantitative analyses show that the algorithm contributes to the performance improvement in presence of path diversity. Simulation results verify our analysis on coding redundancy and buffer size, and also demonstrate that FMTCP improve the performance of multipath TCP when paths have diverse loss and delay.