Multipath data transmission techniques have been proposed to aggregate the resource of multiple heterogeneous access networks to enhance the robustness and throughput of multi-homed mobile devices. The Multipath TCP (MPTCP) protocol has been standardized by the IETF as an extension of conventional TCP, which enables establishing several subflows over multiple interfaces for simultaneous data transmission. However, current implementation of MPTCP only supports static subflow management strategies, which cannot adapt to the changing to network environment. In this paper, we focus on the issues of dynamic subflow management for MPTCP. We present theoretical analysis to the overall throughput of establishing multiple subflows with MPTCP, based on which we propose an adaptive subflow allocation algorithm to determine the proper number of subflows for each network interface to optimize the performance gain. We implement the proposed algorithm, integrate it into the MPTCP Linux Kernel implementation, and evaluate its performance on a wireless network testbed. Extensive experiments show that the proposed allocation algorithm significantly outperforms the existing subflow management strategies with 57% improvement.

已经提出了多径数据传输技术来聚合多个异构接入网络的资源，以增强多归属移动设备的鲁棒性和吞吐量。IETF已将多路径TCP（MPTCP）协议标准化为传统TCP的扩展，它可以在多个接口上建立多个子流以同时进行数据传输。但是，当前MPTCP的实现仅支持静态子流管理策略，无法适应网络环境的变化。在本文中，我们关注于MPTCP的动态子流管理问题。我们对使用MPTCP建立多个子流的总体吞吐量进行理论分析，基于此，我们提出了一种自适应子流分配算法，以确定每个网络接口的适当子流数，以优化性能增益。我们实现了所提出的算法，将其集成到MPTCP Linux内核实现中，并在无线网络测试平台上评估了其性能。大量实验表明，所提出的分配算法明显优于现有的子流管理策略，改进了57％。