Measurement-in-the-Middle

该项目介绍了一种新的测量和分析方法，可用于表征端到端的性能测量（如沿沿着的损耗和延迟）以及内部网络特性（如沿沿着远路径段经历的损耗），从网络内的一个单点进行测量。所提出的“测量中的中间”的方法使用的时间戳，数据包级的痕迹收集在这个单一的测量点与概率状态为基础的发送器/接收器模型和统计推断技术来推断端端的性能措施。这些度量包括沿着源、目的地和测量点之间的路径以及沿着这些路径的部分的损耗和延迟。在这个项目中进行的研究将开发和分析（通过建模，仿真和实证研究）的理论基础和实际考虑需要这种新的方法对网络性能的测量和分析。从网络内的单个点推断端到端和路径部分性能的能力具有许多优点。最重要的是，由于测量是在“中间”进行的，因此可以测量大量不同的源/目的地对的端-端行为，而不必测量单个的发射机和接收机。这项研究将开发一种全新的网络测量方法，使网络运营、工程和研究用户能够测量比过去更大、更多样化的网络流量组合。除了更全面地表征网络流量外，更丰富的测量集和更好地了解网络流量还可以在改进网络管理、网络建模和网络/协议设计方面发挥重要作用。

This project introduces a new measurement and analysis methodology that can be used to characterize end-end performance measures (such as loss and delay along a path) as well as internal network characteristics (such as loss experienced along distant path segments) from measurements taken at a single point within the network. The proposed "measurement-in-the-middle" approach uses time-stamped, packet-level traces gathered at this single measurement point together with probabilistic state-based sender/receiver models and statistical inferencing techniques to infer end-end performance measures. These metrics include the loss and delay along paths between a source, destination, and measurement point, and along portions of these paths as well. The research performed in this project will develop and analyze (through modeling, simulation, and empirical study) the theoretical foundations and practical considerations needed for this new approach towards network performance measurement and analysis. The ability to infer end-to-end and path-portion performance from a single point within the network has many advantages. Most importantly, because measurements are taken "in-the-middle," the end-end behavior of a large and diverse mix of source/destination pairs can be measured, without having to instrument individual senders and receivers. This research will develop a fundamentally new approach to network measurement methodology that will allow network operations, engineering and research users to measure a larger, and more diverse, mix of network traffic than has been possible in the past. In addition to more fully characterizing network traffic, a richer set of measurements and a better understanding of network traffic can also play an important role in improved network management, network modeling, and network/protocol design.