高速无线广域网环境中敏捷和自学习的高效传输方法研究

The increasingly diverse network types (e.g. high-speed wireless networks, cellular mobile networks and datacenter networks) and application requirements make the existing transmission control protocol (TCP) facing the issues of efficiency and reliability etc. Due to the adaptability and predictability, the agile and self-learned transmission methods are attracted gradually. They can use the scarce network bandwidth effectively and achieve the characteristics of low latency, high throughput, low packet loss rate and fairness etc. In this project, we design and implement new transmission methods and frameworks with the characteristics of agile, self-learned, and decoupling the congestion control and flow control in high-speed wireless wide area networks. They can solve the utility issues of TCPs, and accelerate the data delivery and improve the link utilization. The research contents include three parts. Firstly, the relationship between the gradient of transmission performance metrics and the congestion window is built and then the agile TCP is proposed. Secondly, the relation model between the transmission performance metrics and the congestion window is established and then a new protocol, which adopts the statistic and online-learning methods, is proposed to protect the data delivery of cellular networks. Thirdly, decoupling the congestion control and flow control, a distributed and self-learned software-defined control framework is designed and implemented. Furthermore, most of proposed transmission methods are implemented, validated and optimized in the open source system kernel. The expected findings have importantly academic and practical values for improving the quality and ability of the network transmission.

日趋多样化的网络类型(如高速无线网络、蜂窝移动网和数据中心网络)和应用需求使现有传输控制协议面临效率和可靠性等问题。敏捷和自学习的传输控制协议具有自适应性和预测性，能更高效利用网络带宽，实现低时延、高吞吐、低丢包率和公平性等目标。项目立足高速无线广域网环境，设计和实现敏捷、自学习、解耦拥塞控制和流控制的传输控制协议和框架，旨在解决高速无线(时变)广域网中传输控制协议面临的效用问题，加速数据投递速率并改善链路利用率。研究内容包括：建立传输性能指标梯度与拥塞窗口控制的关系，设计和实现敏捷的传输控制协议；建立传输性能指标与拥塞窗口的关系模型，设计和实现统计与在线学习的蜂窝网络性能保障传输控制协议；解耦拥塞控制和流控制，设计和实现分布式自学习的软件定义控制的传输框架；在开源系统内核中实现上述相关传输控制协议，并验证和优化。预期成果对于提升网络传输质量和能力具有重要的学术意义和应用价值。

日趋多样化的网络类型(如高速无线网络、蜂窝移动网和数据中心网络)和应用需求使现有传输控制协议面临效率和可靠性等问题，设计和实现满足不同需求的传输控制协议面临诸多挑战。敏捷和自学习的传输控制协议以其自适应性和预测性逐渐受到关注，能更高效利用网络带宽，实现低时延、高吞吐、低丢包率和公平性等目标。本项目立足高速无线广域网环境，设计和实现敏捷、自学习、解耦拥塞控制和流控制的传输控制协议和框架，旨在解决高速无线广域网中传输控制协议面临的效用问题，加速数据投递速率并改善链路利用率。研究内容包括：设计和实现敏捷的传输控制协议；设计和实现基于统计学习的蜂窝网络性能保障传输控制协议；解耦拥塞控制和流控制，设计和实现分布式的软件定义控制的传输框架；在开源系统内核中实现上述相关传输控制协议，并验证和优化。截止目前，该项目已基本完成项目规定的目标，完成度达90%以上，发表和录用论文9篇(SCI论文4篇、EI论文4篇)，申请发明专利2项，初步开发了传输控制服务的Demo系统(将继续探索)，培养青年教师1名、研究生4名、本科生1名。该项目部分成果的应用对于增强网络信息服务设施的能力具有实际意义，未来深入探索的拥塞控制即服务的系统将解决传输控制协议部署难、环境适应能力差等问题。