NeTS-NR: A Novel Technique to Detect Shared Congestion and Applications

在这个项目中，发现了一种新的技术，可以用来检测任意两条Internet路径之间的共享连接，称为延迟相关小波去噪(DCW)。使用此技术，不需要用于同步的公共端点(源或目标)。更具体地说，DCW的同步偏移高达一秒，足以检测任意两条互联网路径之间的共享拥塞。此外，无论在拥塞点使用随机早期丢卡(RED)还是丢尾排队方法，DCW都能有效地工作。为了将DCW应用于特定的应用，设计了新的协议，指定节点如何生成探测，收集和处理没有公共端点的Internet路径的延迟样本，并向应用提供共享的拥塞检测结果。该协议包括松散同步技术，以将探测序列之间的任何同步偏移限制在最大值。协议是为在树状和网状拓扑的覆盖网络上运行的应用程序设计的，目的是通过消除共享瓶颈来提高应用程序的性能。DCW技术有望实现构建具有改进的拓扑的覆盖网络的协议设计，其中消除了共享瓶颈。DCW的这种应用将有助于设计下一代研究网络和跨学科协作工具，以及共享的研究和教育平台。

Within this project, a novel technique that can be used to detect shared conges-tion between any two Internet paths is discovered, called Delay Correlation with Wavelet denoising (DCW). With this technique, a common end point (source or destination) for synchronization is not needed. More specifically, DCW works efficiently and effectively for a synchronization offset of up to one second, which is large enough for detecting shared congestion between any two Internet paths. Furthermore, DCW works effectively irrespective of whether Random Early Dis-card (RED) or drop-tail queueing method is used at the point of congestion. To apply DCW to specific applications, new protocols are designed that specify how nodes generate probes, collect and process delay samples for Internet paths with no common end point, and provide the shared congestion detection out-come to the applications. The protocols include a loose synchronization tech-nique to limit any synchronization offset between probe sequences to the maxi-mum value. Protocols are designed for applications that run on overlay net-works with both tree and mesh topologies with the objective of improving the applications' performance by removing shared bottlenecks. The DCW technique is expected to enable the design of protocols that construct overlay networks with improved topologies where shared bottlenecks are removed. Such applica-tion of DCW will contribute toward the design of next-generation research net-works and tools for interdisciplinary collaboration, as well as shared research and education platforms.