CRII: NeTS: The Battle for Bandwidth: Heterogeneous Congestion Control on Today's Internet

互联网拥塞越来越频繁地成为家庭网络、互联网服务提供商、企业和云计算中心的挑战。当发生拥塞时，它会导致数据包丢失，增加网络延迟，并降低单个连接的吞吐量-所有这些都会导致在Internet上运行的应用程序性能低下。仔细观察拥塞可以发现，互联网上有赢家和输家：当带宽成为稀缺资源时，一些拥塞会挤出其他拥塞，控制链路容量并损害其他连接的性能。分析这些赢家和输家变得具有挑战性，因为每个互联网服务都可以自由地独立确定其发送速率，根据其拥塞控制算法的参数，加速以消耗更多带宽或减慢以共享链路。常见的拥塞控制（CC）算法包括NewReno、Cubic、Akamai的FastTCP和Google的BBR。本提案旨在了解当前使用的异构拥塞控制算法的公平带宽分配状态。我们的项目目标是双重的：（1）测量当今“在野外”使用的CC算法;以及（2）在与其他算法竞争的受控实验中评估每个算法。为了实现（1），我们将设计和实现新的技术来识别和描述用于任意互联网服务的CC（即使我们自己无法控制服务）。为了实现（2），我们提出了新的框架来进行公平性的科学评估，并在可能的情况下，将我们的公平性测量与描述我们结果的数学模型相结合。该奖项反映了NSF的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

Internet congestion is more and more frequently a challenge in home networks, Internet service providers, enterprises, and cloud datacenters. Where congestion occurs, it introduces packet loss, increases network latency, and slows down throughput for individual connections - all of which contribute to poor performance for applications running over the Internet. A close look at congestion can reveal that there are winners and losers on the Internet: when bandwidth becomes a scarce resource, some senders crowd others out, dominating link capacity and harming the performance of other connections. Analyzing these winners and losers becomes challenging because every Internet service is free to independently determine its sending rate, speeding up to consume more bandwidth or slowing down to share the link, according to the parameters of its congestion control algorithm. Examples of common congestion control (CC) algorithms include NewReno, Cubic, Akamai's FastTCP, and Google's BBR.This proposal aims to understand the state of fair bandwidth allocation given the heterogenous congestion control algorithms in use today. Our project goal is twofold: (1) to measure what CC algorithms are in use 'in the wild' today; and (2) to evaluate each algorithm in controlled experiments in contention with other algorithms. To achieve (1), we will design and implement new techniques to identify and characterize the CC in use for arbitrary Internet services (even when we cannot control the service ourselves). To achieve (2), we propose new frameworks to conduct scientific evaluation of fairness and, where possible, combine our fairness measurements with mathematical models describing our results.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.