基于性能服务的高速网络运输层拥塞控制解决方案

摘要

To improve the congestion control performance in high speed networks, a united framework was proposed for TCP and UDP, the two dominant transport layer protocols. As a layered structure, this framework utilizes the performance service from the upper layer to learn the internal network states. The states information is then used to derive a control objective on the extra packet amounts. The down layer, composed of end systems, achieves that objective by means of window or rate adjustment in reacting to lose or delay signals. Down layer algorithms for TCP and UDP were described. To improve the robustness to measurement errors, fuzzy inference technique was employed in objective deriving. Simulations validate that the proposed schemes are able to achieve high efficiency as well as fairness for both the protocols. Besides , UDP exhibits friendliness and smoothness competing with TCP flows. The system shows its robustness to reasonable measurement errors.%为改善高速网络拥塞控制性能,提出了一种2层结构的运输层解决方案.将TCP和UDP纳入同一个框架,利用上层的性能服务感知内部网络状态,然后在此基础上形成基于冗余分组的下层控制目标.由端系统组成的下层实体通过对时延和丢包的响应以窗口调节或速率调节的方式实现上层设定的公共目标.针对TCP和UDP协议,分别描述了下层端系统执行的拥塞控制算法.另外,为提高系统对测量误差的健壮性,控制目标推导过程使用了模糊推理技术.仿真实验表明,提出的拥塞控制方案,与TCP流竞争带宽的UDP流在达到TCP友好的同时保持了平滑的吞吐率输出,并且系统对合理范围内的测量误差表现出较强的健壮性.