【作者】 王勇；

【导师】 阳旺；

【作者基本信息】 中南大学 ， 电子信息（专业学位）， 2024， 硕士

【摘要】 近年来,随着新兴应用的发展,用户对于网络传输的吞吐量和时延提出了更高的要求。命名数据网络(Named Data Networking,NDN)是以内容分发为初衷的网络架构,因为多源多路径的特性,有希望为新兴应用提供更好的网络支持。然而,在NDN中,随着源和路由节点的增加,用户可用子路径的数量显著增加,子路径间交叠的可能性增加,共享瓶颈也变得更为普遍。共享瓶颈的存在为多路径传输机制的设计带来了挑战,并且将直接影响网络性能。因此,本文首先提出了共享瓶颈检测算法(CDDTW-SBD)。在此基础上,本文围绕着共享瓶颈,对多路径选择和多路径拥塞控制进行了优化,并进一步提出了一种共享瓶颈感知的多路径传输机制(SBAMPT)。本文主要工作如下:(1)针对路径滞后带来的检测结果不准确的问题,本文提出基于拥塞度和动态时间规整的共享瓶颈检测算法。该算法定义子路径中间节点的最大排队比率为子路径拥塞度,接着基于动态时间规整算法进行子路径间的拥塞度相似性计算,从而最终判断他们是否共享瓶颈。算法中通过构建瓶颈集的方式保存瓶颈检测的结果,并且以同源瓶颈集为初始瓶颈集以加速检测的过程。除此之外,算法还加入了剪枝的过程以减少计算量。(2)针对共享瓶颈带来的网络性能下降和缺乏细粒度的拥塞控制的问题,本文在检测共享瓶颈的基础上,进一步围绕共享瓶颈设计多路径传输机制。该机制中的多路径选择主要基于瓶颈集信息和子路径的质量测量,在每一个瓶颈集上启发式地搜索能够充分利用瓶颈带宽的最优子路径组合,从而减轻共享瓶颈的影响。在多路径拥塞控制方面,该机制还补充了基于子路径拥塞度的拥塞窗口调整步幅设计和基于瓶颈集中子路径数量的窗口下降时机设计。实验结果表明,在存在共享瓶颈的场景中,本文提出的共享瓶颈检测算法CDDTW-SBD能够有效减轻路径滞后的干扰,提高了约26%的检测精度。SBAMPT中关于多路径选择和多路径拥塞控制的设计都有效降低了时延。在最终多路径传输机制的评估中,SBAMPT相比于PCON提高了约24.6%的平均吞吐量,降低了约50%的平均时延。相比于MPCC,SBAMPT在不损失吞吐量的前提下降低了约58%的平均时延。图27幅,表2个,参考文献61篇更多还原

【Abstract】 In recent years,with the development of emerging applications,users are demanding higher throughput and latency for network transmis-sion.Named Data Networking(NDN),a network architecture with the orig-inal intention of content distribution,has the promise of providing better network support for emerging applications because of its multi-source and multi-path nature.However,in NDN,as the number of source and routing nodes in-creases,the number of sub-paths available to users increases significantly,the possibility of overlapping between sub-paths increases,and shared bot-tlenecks become more prevalent.The existence of shared bottlenecks cre-ates challenges for the design of multi-path transmission mechanisms and will directly affect network performance.Therefore,this thesis first pro-poses a shared bottleneck detection algorithm(CDDTW-SBD).Based on this,this thesis optimises multi-path selection and multi-path congestion control around shared bottlenecks,and further proposes a shared bottleneck aware multi-path transmission mechanism(SBAMPT).The main work of this thesis is as follows:(1)Aiming at the problem of inaccurate detection results caused by path lag,this thesis proposes a shared bottleneck detection algorithm based on congestion degree and dynamic time warping.The algorithm defines the maximum queuing ratio of the intermediate nodes on the sub-paths as the sub-path congestion degree,and then calculates the congestion degree similarity between the sub-paths based on the dynamic time warping algo-rithm,so as to determine whether they share the bottleneck or not.The algorithm saves the results of bottleneck detection by constructing a bottle-neck set,and uses the homologous bottleneck set as the initial bottleneck set to speed up the detection process.In addition,the algorithm also adds the pruning process to reduce the amount of computation.(2)Aiming at the network performance degradation caused by shared bottlenecks and the lack of fine-grained congestion control,this thesis fur-ther designs a multi-path transmission mechanism around shared bottle-necks based on the detection of shared bottlenecks.The multi-path selection in this mechanism is mainly based on the bottleneck set information and the quality measurement of sub-paths,and heuristically searches for the opti-mal sub-path combinations that can fully utilise the bottleneck bandwidth on each bottleneck set,so as to mitigate the impact of shared bottlenecks.For multi-path congestion control,the mechanism is complemented by a congestion window adjustment step design based on sub-path congestion and a window descent timing design based on the number of sub-paths in the bottleneck set.Experimental results show that in scenarios where shared bottlenecks exist,the shared bottleneck detection algorithm CDDTW-SBD proposed in this thesis is effective enough to mitigate the interference of path lag and improve the detection accuracy by about 26%.The designs of SBAMPT on both multi-path selection and multi-path congestion control are effective in reducing the latency.In the evaluation of the final multi-path transmission mechanism,SBAMPT improves the average throughput by about 24.6% and reduces the average delay by about 50% compared to PCON.Compared to MPCC,SBAMPT reduces the average delay by about 58% without loss of throughput.更多还原