NeTS: Small: A Principled Approach to Enabling Policy Transparency for Mobile Networks

诸如视频流之类的高带宽应用的流行以及资源受限的移动的网络的流行重新引发了关于互联网服务提供商（ISP）如何处理与不同应用相关联的网络流量的讨论。为了管理稀缺的网络资源，ISP使用中间盒来选择性地实施网络策略（例如，限制视频流的传输速率）。本研究将探讨中间盒对网络流量的影响，并引导网络用户、政策制定者和监管者更深入地理解运营网络中部署的策略。本项目解决了限制研究人员理解中间盒施加的网络策略的三个关键挑战。首先，研究人员很难描述政策的特征，因为他们只能有限地或根本无法访问中间盒进行实验。其次，研究人员缺乏中间盒如何分类和与网络流量交互的模型，因此很难对它们进行推理。最后，识别和理解中间盒行为通常需要从网络内部的Vantage位置进行测量。这项调查需要对商业中间盒进行黑盒和灰盒实验，以建立它们如何运作的准确模型，并使用这些观察结果来建立一个可扩展的，灵活的和现实的测试平台，用于评估检测和隔离网络策略的方法。该项目将结合基于跟踪的分析来设计高效，弹性的算法来检测终端系统的这些属性，并优化该检测系统的规模。为了推广我们的结果，PI将开发一种理论，说明中间盒如何影响流量，如何从端点检测这些中间盒，可能对网络的可见性有限。

The rise in popularity of bandwidth-hungry applications such as video streaming, and the popularity of resource-constrained mobile networks has reignited discussions about how network traffic associated with different applications is treated by Internet Service Providers (ISPs). To manage scarce network resources, ISPs use middleboxes to selectively enforce network policies (e.g., limit transfer rates for video streaming). This study will explore the impact of middleboxes on network traffic, and lead network users, policymakers, and regulators to a deeper understanding of policies deployed in operational networks.This project addresses three key challenges that limit researchers' understanding of network policies imposed by middleboxes. First, researchers struggle to characterize policies because they have limited or no access to middleboxes for experimentation. Second, researchers lack models of how middleboxes classify and interact with network traffic, making it difficult to reason about them. Finally, identifying and understanding middlebox behavior often requires measurements from vantage points inside the network. This investigation entails black-box and grey-box experiments on commercial middleboxes to build accurate models of how they operate, and using these observations to build a scalable, flexible and realistic testbed for evaluating methods to detect and isolate network policies. This project will incorporate trace-based analysis to design efficient, resilient algorithms to detect these properties from end systems, and optimizations to make this detection system scale. To generalize our results, the PIs will develop a theory of how middleboxes can impact traffic, how to detect these middleboxes from endpoints, potentially with limited visibility into the network.