An investigation of Starlink's performance during the May'24 solar superstorm

Low Earth Orbit (LEO) satellites have revolutionized the consumergrade Internet market. The main giant of this landscape, Starlink, is already operating the world’s largest LEO satellite fleet of 8,000 satellites made of non-radiation-hardened components. The recent May 2024 solar superstorm created an opportunity to evaluate the performance and reliability of such a network under intense solar events. In this paper, we conduct a statistical study on the packet loss, latency, and orbital drag experienced by satellites from a longterm perspective. The results indicate marginal inflation in loss and latency during and immediately after the superstorm. While increasing the observation window size dilutes the inflation under regular performance fluctuations. Additionally, we list out a few roadblocks that need to be addressed to pinpoint the impact on any specific satellite, along with the end-user’s network connectivity experience caused by solar radiation.

低地球轨道（LEO）卫星已经彻底改变了消费级互联网市场。该领域的主要巨头——星链（Starlink），目前运营着全球最大的LEO卫星舰队，其8000颗卫星由非抗辐射加固组件构成。最近的2024年5月太阳超级风暴为评估此类网络在强烈太阳事件下的性能和可靠性创造了机会。在本文中，我们从长期视角对卫星经历的丢包、延迟和轨道拖曳进行了统计研究。

研究结果表明，在超级风暴期间及紧随其后，丢包和延迟出现了轻微的上涨。然而，当扩大观测窗口时，这种上涨效应在常规的性能波动下被稀释了。

此外，我们列出了一些在精准定位太阳辐射对特定卫星的影响，以及由此对终端用户网络连接体验造成影响的研究中需要解决的障碍。

Introduction

Space weather is a menace to space-borne electronics. Solar flares [23, 48] and Coronal Mass Ejections (CMEs) [20, 53] from the Sun’s active regions can release extreme ultraviolet (EUV) radiation [14, 52], X-Rays [54], or a burst of dense, hot, high-velocity plasma into outer space. This surge of plasma consists of strong magnetic fields; hence, the interaction with the Earth’s magnetic field leads to geomagnetic storms [27, 32]. Exposure to radiation from highly charged particles presents a constant risk to manmade infrastructure in space. There is a documented history of onboard instruments and even complete mission failures due to these intense space weather events [18, 25, 30, 36, 44]. However, the lack of sufficient data points has always been a challenge in studying the effects of past solar events.

The 21st century’s advancements in space technology, i.e., the reduction in launch cost and production of small satellites with relatively low-cost commodity hardware [5, 19, 22], have enabled SpaceX’s Starlink (and many others) to deploy a massive constellation of thousands of LEO satellites. Their success in serving worldwide 5 million customers [12] under a global ISP has now opened an opportunity to investigate the LEO operational challenge due to solar events and its implications on the Internet service experience. Many researchers have already started utilizing this opportunity to explore areas such as the failure of the 38 Starlink satellite deployment [15, 17] and the impacts on the satellite’s orbital stability due to solar events [16]. However, a significant gap still exists in quantifying the effect of the degradation of satellite link quality on end-user Internet experience and pinpointing the cause-and-effect relationship from solar events to particular infrastructure.

After last year’s, i.e., May 2024 solar superstorm, an effort [47] has been made to quantify the impact of such an event on Starlink’s LEO network using 81 RIPE Atlas probes deployed across 18 countries. Their analysis illustrated an immediate rise in packet loss and a delayed inflation in round-trip time (RTT) after the commencement of the superstorm. While Starlink, in their response to the FCC public notice seeking comment on the impact of this G5 class storm [21], stated that the service experienced less than 1 minute of disruption and continued without degradation [26]. They praised the advanced collision avoidance and auto station-keeping system, which kicked in real-time with capable thrusters to countermeasure against 2 − 5× orbital drag between altitudes 300 to 550 km during the event.

Given that prior work [47] has focused on the Starlink performance within a 15 day window of the superstorm, we broaden this window of analysis to draw a conclusion from a long-term perspective. A statistical analysis of Starlink performance measurements over a couple of months shows a negligible difference in performance measurements during this superstorm as compared to the long-term performance characteristics. Shortening this window close to the May 2024 geomagnetic storm does show a marginal inflation on packet loss and a barely visible increase in latency. The magnitudes of inflation are not large enough to be concerned about, as they remain within the range of regular performance fluctuations that occur over Starlink networks. Additionally, we experience and state some limitations of existing satellite link measurements to deep dive into such studies to establish a concrete causality from solar radiation to impact on LEO infrastructure, and then user-perceived Internet performance implications at the user end. Given that we have already started approaching the minima of the current solar cycle, we are left with a short window to address these issues in the current cycle.

The remaining paper is organized as follows – §2 discusses the background and related works. Then §3 discusses the dataset used for the analysis. In §4, we investigate the performance implications of the May ′ 24 superstorm. Finally, the paper is concluded in §5.

空间天气对星载电子设备是一种威胁。来自太阳活动区的太阳耀斑 [23, 48] 和日冕物质抛射（CME） [20, 53] 会向外太空释放极紫外（EUV）辐射 [14, 52]、X射线 [54]，或是一股稠密、炽热、高速的等离子体爆发。这股等离子体流包含强磁场，因此其与地球磁场的相互作用会导致地磁暴 [27, 32]。暴露于高能带电粒子的辐射对太空中的人造基础设施构成了持续的风险。历史上有记载的因这些强烈的空间天气事件导致星上仪器乃至整个任务失败的案例 [18, 25, 30, 36, 44]。然而，缺乏足够的数据点一直是研究过去太阳事件影响的一个挑战。

21世纪空间技术的进步，即发射成本的降低和采用相对低成本商用硬件的小型卫星的生产 [5, 19, 22]，使得SpaceX的星链（以及其他许多公司）能够部署由数千颗LEO卫星组成的庞大星座。他们作为全球性互联网服务提供商（ISP）成功服务全球500万客户 [12] 的事实，为我们研究因太阳事件带来的LEO运营挑战及其对互联网服务体验的影响提供了机会。许多研究人员已经开始利用这一机会探索诸如38颗星链卫星部署失败 [15, 17] 以及太阳事件对卫星轨道稳定性的影响 [16] 等领域。然而，在量化卫星链路质量下降对终端用户互联网体验的影响，以及精准定位从太阳事件到特定基础设施的因果关系方面，仍存在显著的研究空白。

在去年的2024年5月太阳超级风暴之后，一项研究 [47] 尝试使用部署在18个国家的81个RIPE Atlas探针来量化此类事件对星链LEO网络的影响。他们的分析显示，在超级风暴开始后，丢包率立即上升，而往返时间（RTT）则出现了延迟性的增加。

然而，星链在其对FCC就此次G5级风暴影响征求意见的公开通知的回应 [21] 中表示，其服务中断时间不到1分钟，并且服务质量未受影响 [26]。他们称赞了其先进的碰撞规避和自动位置保持系统，该系统在事件期间通过强大的推进器实时启动，以对抗海拔300至550公里之间2至5倍的轨道拖曳。

鉴于先前的工作 [47] 聚焦于超级风暴前后15天的时间窗口，我们拓宽了这一分析窗口，以便从更长期的视角得出结论。对星链几个月性能测量的统计分析显示，与长期性能特征相比，这次超级风暴期间的性能测量差异可以忽略不计。当将窗口缩短至接近2024年5月地磁暴时，确实显示出丢包率的轻微上涨和几乎不可见的延迟增加。但这些上涨的幅度不足为虑，因为它们仍在星链网络日常性能波动的范围内。此外，我们指出了现有卫星链路测量在深入此类研究以建立从太阳辐射到LEO基础设施影响，再到用户感知的互联网性能影响这一明确因果链时所面临的一些局限性。考虑到我们已开始接近当前太阳活动周期的极小期，我们在本周期内解决这些问题的时间窗口已经很短。

本文其余部分的组织如下：第2节讨论背景和相关工作。第3节讨论用于分析的数据集。在第4节中，我们调查了24年5月超级风暴的性能影响。最后，第5节对本文进行总结。

Background and related work

This section provides an overview of the May 2024 geomagnetic superstorm and previous research.

本节概述2024年5月的地磁超级风暴及先前的研究。

May 2024 solar storm: Between May 8 and 10, 2024, the Sun released six X-class solar flares, including one X3.98 class flare on May 10 from the NOAA active region (AR13664) [34]. These events are reflected as spikes in X-Ray flux captured by the NOAA GOES satellite at geostationary orbit [41], as shown in Fig. 1 (top). These back-to-back eruptions of plasma or CMEs collided and merged with each other, forming a composite solar wind consisting of a strong magnetic field while moving toward Earth [49]. The impact of this solar wind was measured by the DSCOVR satellite at L1 1 [40], showing sudden increases in the velocity of charged particles, along with varying magnetic fields after May 10, 2024, at 16:30 UTC, as shown in Fig. 1. Approximately at 17:00 UTC on May 10, this gust of solar wind collided with the Earth’s magnetosphere. The effects on the magnetosphere are reflected in the Dst index, which fell to −412 nanoTesla (nT) by May 11 at 02:00 UTC. According to the NOAA space weather scale [42], this is classified as a G5 class, an extreme geomagnetic storm. This is the strongest storm recorded after two decades, since the 2003 Halloween solar storm [1, 7]. These intense space weather conditions compressed the Earth’s dayside magnetosphere (magnetopause), pushing it below the geostationary orbit (35,786 km from Earth) for several hours. Thus, magnetic field disturbances during this event were observed by the NOAA GOES18 geostationary satellite [39], as shown in Fig. 1 (bottom).

2024年5月太阳风暴： 在2024年5月8日至10日期间，太阳从NOAA活动区（AR13664）释放了六次X级太阳耀斑，其中包括5月10日的一次X3.98级耀斑 [34]。这些事件在NOAA GOES卫星于地球静止轨道上捕捉到的X射线通量中表现为尖峰 [41]，如图1（顶部）所示。这些接连爆发的等离子体或CME在向地球移动的过程中相互碰撞并合并，形成了一股包含强磁场的复合太阳风 [49]。这股太阳风的影响由位于L1点的DSCOVR卫星测得 [40]，显示在2024年5月10日16:30 UTC之后，带电粒子速度突然增加，磁场也随之变化，如图1所示。大约在5月10日17:00 UTC，这股太阳风与地球磁层相撞。对磁层的影响反映在Dst指数上，该指数到5月11日02:00 UTC时已降至-412纳特斯拉（nT）。根据NOAA空间天气等级 [42]，这被归类为G5级，即极端地磁暴。这是自2003年万圣节太阳风暴 [1, 7] 以来记录到的最强风暴。这些强烈的空间天气条件压缩了地球的日侧磁层（磁层顶），使其在数小时内被推至地球静止轨道（距地球35,786公里）以下。因此，NOAA GOES18地球静止卫星 [39] 在此事件期间观测到了磁场扰动，如图1（底部）所示。

Related work: The impact of space weather or geomagnetic storms on Earth has always been an active area of research. Authors in [16, 37, 43] have explored the satellite’s orbital drag, particularly in LEO during solar events. Some works [15, 17] have investigated the loss of 38 Starlink satellites in February 2022 due to a minor geomagnetic storm. More recently, following the May 2024 geomagnetic superstorm, the authors in [45] have discussed the characteristics of satellite drag and decay during the event, while in [14], the authors have studied how the preconditioning of the superstorm could have led to the early reentry of 12 decommissioned Starlink satellites. Authors in [49, 55] have unveiled the impact of the event on Earth’s magnetosphere and ionosphere.

相关工作： 空间天气或地磁暴对地球的影响一直是一个活跃的研究领域。文献 [16, 37, 43] 的作者探讨了太阳事件期间卫星的轨道拖曳，特别是在LEO轨道。一些工作 [15, 17] 调查了2022年2月因一次小型地磁暴导致38颗星链卫星损失的事件。最近，在2024年5月地磁超级风暴之后，文献 [45] 的作者讨论了事件期间卫星拖曳和衰减的特征，而文献 [14] 的作者研究了超级风暴的预处理如何可能导致12颗已退役的星链卫星提前再入大气层。文献 [49, 55] 的作者揭示了该事件对地球磁层和电离层的影响。

However, the majority of the works [14–17, 37, 43, 45] have studied the physical impacts of geomagnetic storms on satellites. On the other hand, the studies in [29, 31, 33, 35, 38, 46] have reported Starlink’s performance measurements under regular weather conditions. Only a recent study [47] has explored Starlink’s network performance during the May 2024 superstorm. Hence, the knowledge of how these space weather events affect the Internet connectivity of LEO satellite networks is quite limited.

Our work: While the prior work [47] zooms into a window of 15 days of May 2024 solar superstorm, in this work, we broaden this window to showcase how Starlink connectivity performance differs when compared against long-term performance characteristics.

然而，大部分工作 [14–17, 37, 43, 45] 研究的是地磁暴对卫星的物理影响。另一方面，研究 [29, 31, 33, 35, 38, 46] 报告了星链在常规天气条件下的性能测量。只有一项最近的研究 [47] 探讨了星链在2024年5月超级风暴期间的网络性能。因此，关于这些空间天气事件如何影响LEO卫星网络互联网连接的知识相当有限。

我们的工作： 先前的工作 [47] 聚焦于2024年5月太阳超级风暴的15天窗口，而在本工作中，我们拓宽了这一窗口，以展示星链的连接性能与长期性能特征相比有何不同。

Preparing the datasets

Acquiring: We use the following datasets for this analysis.

Network measurement: We rely upon RIPE Atlas probes [9] connected to Starlink. We use the RIPE Atlas REST API [10] to acquire details of all the 96 probes across 24 countries attached to Starlink’s Autonomous System Number (ASN 14593 [8]). Then, we fetch 1,551 (source and destination pairs) built-in periodic ping measurements over IPv4 and IPv6 from April 1 to May 31, 2024.

Orbital drag: We use CosmicDance [16] to acquire drag term [28] from the publicly available TLEs of Starlink’s satellites.

Cleaning: Next, to minimize the inaccuracy in the analysis, we take the following steps to clean up the acquired datasets.

Network outages: We explore the public domain reports on network issues and remove all the measurements data after May 28, 2024, since Starlink global outage is reported by CISCO ThousandEyes on May 29, 2024 [51].

Socioeconomic factor: We also remove all the probes from Ukraine, due to reports of electronic warfare and jamming [24].

Data anomalies: In the remaining datasets, we found that the RTT between many source-destination pairs is not stable and shifts at arbitrary levels for a duration of days, irrespective of the intensity of solar events, as shown in Fig. 2(a) with RTT measurements in probe 1004232. We speculate that this is because of the frequent changes in the assigned Points of Presence (PoP) [29, 38]; however, we do not find any clear evidence because we are missing the route traces. To tackle this, we adopt a simple yet effective method: calculating the median of consecutive RTT points in the timeseries window of length 2,000 (empirically decided), which is denoted with a red dot in Fig. 2. The difference between the maximum and minimum of the median values (we call this ‘stability score’), in Fig. 2(a) is 169 ms, which is much higher than the 2.57 ms in a stable RTT measurement over two months in Fig. 2(b). We use this stability score throughout the remaining paper to analyze the performance.

数据获取： 我们使用以下数据集进行分析。

• 网络测量：我们依赖连接到星链的RIPE Atlas探针 [9]。我们使用RIPE Atlas REST API [10] 获取了连接到星链自治系统号（ASN 14593 [8]）的24个国家共96个探针的详细信息。然后，我们获取了从2024年4月1日至5月31日期间，1551个（源和目标对）基于IPv4和IPv6的内置周期性ping测量数据。
• 轨道拖曳：我们使用CosmicDance [16] 从公开的星链卫星双行轨道根数（TLE）中获取拖曳项 [28]。

数据清洗：接下来，为了最小化分析中的不准确性，我们采取以下步骤清理获取的数据集。

• 网络中断：我们查阅了关于网络问题的公开报告，并移除了2024年5月28日之后的所有测量数据，因为思科ThousandEyes报告了2024年5月29日的星链全球中断事件 [51]。
• 社会经济因素：由于有关电子战和干扰的报告 [24]，我们还移除了所有来自乌克兰的探针数据。

数据异常： 在余下的数据集中，我们发现许多源-目标对之间的RTT不稳定，并且会在任意时间点发生持续数天的阶梯式变化，这与太阳事件的强度无关，如图2(a)中探针1004232的RTT测量所示。我们推测这是由于分配的接入点（PoP） [29, 38] 频繁变更所致；然而，由于缺少路由追踪数据，我们没有找到明确的证据。为了解决这个问题，我们采用了一种简单而有效的方法：在长度为2000（经验性决定）的时间序列窗口中，计算连续RTT点的中位数，这在图2中用红点表示。图2(a)中中位数的最大值和最小值之差（我们称之为“稳定性得分”）为169ms，远高于图2(b)中稳定RTT测量两个月内的2.57ms。我们在本文的其余部分使用这个稳定性得分来分析性能。