DISCUSSIONS AND OPEN QUESTIONS¶
Generality. Experiments in this paper were mainly conducted based on Starlink, since currently it is the only operational broadband LSN available for individual users. Since LEO mobility is an inherent characteristic of all LSNs, we believe all LSN operators need a reconfiguration-like mechanism to manage and schedule their ground-to-space connections, which can assist end-to-end congestion control. We will extend our study to other operational LSNs in the future.
通用性
本文的实验主要基于Starlink进行，这是目前唯一面向个人用户可用的运营宽带低地轨道卫星网络（LSN）。由于LEO卫星的移动性是所有LSN的固有特性，我们相信所有LSN运营商都需要类似重新配置的机制来管理和调度其地面到空间的连接，这对于端到端拥塞控制具有重要意义。未来，我们将把研究扩展到其他运营中的LSN。
Limitations of existing methods for obtaining reconfiguration information. Based on the gRPC service provided by the Starlink terminal, an end host can detect whether a reconfiguration occurs in seconds. However, this time granularity may not be sufficient to ultimately achieve accurate congestion control that can adapt to millisecond-level network changes. We leave the exploration of a practical and more accurate method to detect LEO reconfiguration and improve real-time congestion control as our future work.
现有方法在获取重新配置信息方面的局限性
基于Starlink终端提供的gRPC服务，终端主机可以在秒级时间粒度内检测是否发生了重新配置。然而，这种时间粒度可能不足以实现适应毫秒级网络变化的精确拥塞控制。我们将探索一种更实用且更精确的方法来检测LEO重新配置，并改进实时拥塞控制，这将作为未来工作的方向。
Finally, we discuss several open questions and hope they can inspire future CCA research for LSNs.
最后，我们讨论了几个开放性问题，并希望它们能够激发未来针对LSN的拥塞控制算法（CCA）研究。
Q1: Considering the drastic, multi-dimensional network variations induced by LEO mobility, are singlebasis CCAs suitable for LSNs? Many CCAs only rely on a single basis to infer network congestion. For example, Cubic depends on loss and Copa/Vegas depends on RTT changes. However, due to the LEO mobility, end-to-end connections can experience various network changes (i.e., capacity, delay and loss) unrelated to congestion. We suspect it is impossible to achieve prompt and accurate congestion detection via only monitoring a single network metric in LSNs, unless there is an explicit signal for congestion discrimination.
Q2: Should LSN operators expose more low-layer information to CCAs? A classic direction to improve CCAs under time-varying network conditions is to use the underlying information that can reflect the channel quality more accurately and promptly to assist CCAs (e.g., [21, 52]). This direction might also work for LSNs, but it requires satellite operators expose sufficient information and APIs on their terminals, but may involve new challenges in system stability and security, if the user has been granted higher privileges.
Q3: Would a unified network monitor tailored for LSNs help? Since all CCAs rely on monitoring certain network metrics, we leave a unified network monitor tailored for LSNs which can characterize rapidly varying LSN conditions more efficiently (e.g., by exploiting those insights in §5) and benefit all CCAs running on the end host as our future work.
Q1: 鉴于LEO移动性引发的多维网络剧烈变化，单一基础的CCA是否适用于LSN？
许多拥塞控制算法（CCAs）仅依赖单一基础来推断网络拥塞。例如，Cubic依赖丢包，Copa/Vegas依赖RTT变化。然而，由于LEO卫星的移动性，端到端连接可能会经历与拥塞无关的各种网络变化（如容量、延迟和丢包）。我们怀疑，仅通过监测单一网络指标无法在LSN中实现快速且准确的拥塞检测，除非存在区分拥塞的显式信号。
Note
我们需要多元化指标，个人感觉可以类比ECMP提出时考虑设定的五元组函数
Q2: LSN运营商是否应向CCAs公开更多低层信息？
在时变网络条件下改进CCAs的一个经典方向是利用能够更准确、及时反映信道质量的底层信息来辅助CCAs（例如，[21, 52]）。这一方向可能也适用于LSN，但它要求卫星运营商在其终端上公开足够的信息和API。然而，如果用户被授予更高权限，这可能会带来系统稳定性和安全性的新挑战。
Note
如果LSNs能够提供更多的运行性能指标（底层数据），那将更有利于上层设计
个人更倾向于这是一种技术与商业行为做balance的倡议
Q3: 针对LSN定制的统一网络监控是否有帮助？
由于所有CCAs都依赖于监控某些网络指标，我们将研究一种针对LSN定制的统一网络监控作为未来工作。该监控工具可以更高效地表征快速变化的LSN条件（例如，利用§5中的洞见），并惠及运行在终端主机上的所有CCAs。
Note
我们会做：针对LSN定制的统一网络监控