RELATED WORK

Internet congestion control has been one of the most active areas of research in computer networking over the past thirty years. Previous efforts closely related to our work include:

互联网拥塞控制是过去三十年来计算机网络研究中最活跃的领域之一。与本文工作密切相关的先前研究包括：

End-to-end congestion control. In addition to those classic CCAs introduced in §3, other previous efforts have studied CCAs specially for rapidly varying networks such as WiFi/cellular networks [30, 31, 33, 39, 50–52]. However, these works rely on specific characteristics (e.g., resource block information in cellular networks) that are unavailable in current LSNs. In addition, [14] depends on low-layer information, which limits its applicability since in current operational LSNs like Starlink, the Link-layer or PHY-layer information of satellite links is not exposed to the end host.

端到端拥塞控制

除了§3中介绍的经典拥塞控制算法（CCA），其他研究还探讨了专门针对快速变化网络（如WiFi/蜂窝网络）的CCA [30, 31, 33, 39, 50–52]。然而，这些研究依赖于某些特定特性（例如蜂窝网络中的资源块信息），这些特性在当前的低地轨道卫星网络（LSN）中不可用。此外，文献[14]依赖于低层信息，这限制了其适用性，因为在当前运营的LSN（如Starlink）中，卫星链路的链路层或物理层信息并未向终端主机公开。

Explicit congestion control. Many previous works propose to use explicit congestion signal to improve CCAs. For example, BMCC [40, 41] introduces explicit control to insert ECN bits in the intermediate nodes with congestion. The recent ABC [21] is an explicit CCA for network paths with time-varying wireless links. ABC requires modifications on the access point (e.g., a WiFi router or LTE base station). Since the satellite access points in an LSN are hard to modify, it is difficult to directly apply ABC in current LSNs.

显式拥塞控制

许多先前的工作提出使用显式拥塞信号来改进CCA。例如，BMCC [40, 41]引入显式控制，在发生拥塞的中间节点插入ECN位。最近的ABC [21]是一种针对时间变化无线链路路径的显式CCA。ABC需要对接入点（如WiFi路由器或LTE基站）进行修改。然而，由于LSN中的卫星接入点难以修改，直接在当前LSN中应用ABC存在困难。

Measuring satellite networks. Authors of [16] compared the performance of TCP Cubic and BBR in a commercial GEO satellite network. The rapid development of recent LSNs has stimulated a number of measurement studies to reveal the architecture [38, 42, 46], physical-layer characteristic [19, 25], network performance [20, 24, 26, 28, 3436, 47, 48, 54], and the impact on social medias [44, 45] of LSNs. Authors of [1, 12] have conducted preliminary measurement studies on classic TCP congestion control in Starlink network, but they did not examine the performance of the latest CCAs (e.g., BBRv3, Copa and VIVACE) and perform a white-box analysis to reveal the root causes (as §3.3). These pioneering measurement efforts complement our work.

卫星网络测量

文献[16]作者比较了TCP Cubic和BBR在商业地球同步轨道（GEO）卫星网络中的性能。最近LSN的快速发展激发了一系列测量研究，以揭示其架构[38, 42, 46]、物理层特性[19, 25]、网络性能[20, 24, 26, 28, 34-36, 47, 48, 54]以及对社交媒体的影响[44, 45]。文献[1, 12]的作者对Starlink网络中的经典TCP拥塞控制进行了初步测量研究，但他们并未考察最新CCA（如BBRv3、Copa和VIVACE）的性能，也未进行白盒分析以揭示根本原因（如§3.3所述）。这些开创性的测量工作补充了本文研究。

Congestion control in ad-hoc networks. In the early 2000s, several modifications like TCP-F [15] and ATCP [32] were introduced to adapt TCP to ad-hoc networks, where the network infrastructure might also be dynamically updated. Several approaches [18, 22, 43] focused on adapting RTO value to enhance TCP performance in ad-hoc networks. Different from traditional ad-hoc networks, operational LSNs have many new features such as the mega-constellation scale, very frequent network dynamics, the predictability of global reconfiguration, and thus involve new challenges on CCAs.

自组织网络中的拥塞控制

在2000年代初期，一些修改方案如TCP-F [15]和ATCP [32]被引入，以适应自组织网络中的TCP，在这种网络中，基础设施可能会动态更新。一些方法[18, 22, 43]专注于调整重传超时（RTO）值以增强TCP性能。与传统自组织网络不同，运营中的LSN具有许多新特性，例如大规模星座、高频率网络动态性、全球重新配置的可预测性，因此在CCA方面面临新的挑战。

Tip

ad-hoc: 特别指定的

ad-hoc network: 自组织网络