Digest · Measuring Starlink · read after Chapter 3
導讀 · 量測 Starlink · 第 3 章後閱讀
Network Characteristics of LEO Satellite Constellations: A Starlink-Based Measurement from End UsersLEO 衛星星座的網路特性：從終端使用者出發的 Starlink 量測
This digest explains how rain, a geomagnetic storm and vehicle motion affect separate parts of Starlink’s access path. It also revisits the Vancouver study’s interpretations using later measurements.讀完後，你能說明一項溫哥華地區研究所量到的雨、地磁風暴與行進中的汽車各自對 Starlink 鏈路造成什麼影響，為什麼這三種效應作用在系統的相異部位，以及論文 2022 年的哪些觀察已被後續研究重新詮釋。
Main result
主要結果
The Vancouver study compares rain, a geomagnetic storm and vehicle motion against a stationary Starlink baseline. Each event stresses a separate part of the access system.
溫哥華研究以靜止 Starlink 為基準，比較雨、地磁風暴與車輛移動，各種事件影響了接入系統的各個部位。
| Item項目 | Values數值 | Conditions and source條件與來源 |
|---|---|---|
| Latency comparison | 40–70 km | Ma et al. compare Vancouver-area Starlink with cable using latency, loss and TCP/UDP tests to nine cloud regions. A drive test covers 40–70 km/h. |
| Evidence and scope | 27% | and the paper's lasting contribution is to show how the environment enters the link: precipitation cut throughput by about 27% on average |
| Capacity and rate | 215 Mb/s · 120 Mb/s · 4.1–5.2 mm · 3–4 · 2022 · 100 · 5 Mb/s | UDP download falls from about 215 to 120 Mb/s under 4.1–5.2 mm rain. During the 3–4 February 2022 storm, download falls from about 100 to 5 Mb/s while upload stays largely intact. |
| Evidence and scope | 16.5 s · 5 s | and in motion the dish lost service at a median interval of 16.5 s for a median of 5 s [53, §IV-D, VI-B]. |
| Latency comparison | 80 Mb/s · 7 Mb/s · 200 · 10 Mb/s · 50 ms · 10% | Stationary TCP averages about 80 Mb/s down and 7 Mb/s up, while UDP reaches about 200 and 10 Mb/s. Latency is around 50 ms, roughly 10% above cable [53, §IV-A, Tab. II]. |
| Evidence and scope | 15 · 17% | The study is one region, one year and one generation of the network, it predates the discovery of Starlink's 15-second scheduler. Its rain figure is larger than the 17% a later two-site preprint found. Thus, it should be quoted with its place and date. |
| 容量與速率 | 40–70 km · 27% · 215 Mb/s · 4.1–5.2 mm · 120 Mb/s · 2022 · 2 · 3–4 · 100 · 5 Mb/s · 16.5 s · 5 s | Ma et al. 在溫哥華地區以地面有線網路為對照量測 Starlink，對九個地區的雲端伺服器執行延遲、遺失與 TCP/UDP 吞吐量測試，並加上時速 40–70 km 的行車測試，論文持久的貢獻是展示環境如何進入鏈路：降水使吞吐量平均下降約 27%，UDP 下載從乾燥天氣的約 215 Mb/s 降到 4.1–5.2 mm 雨量時的約 120 Mb/s，2022 年 2 月 3–4 日的地磁風暴使下載從約 100 降到 5 Mb/s，上傳則大致保持固定，而在移動中終端以中位數 16.5 s 的間隔失去服務，中位數持續 5 s[53, §IV-D, VI-B]。 |
| 容量與速率 | 80 Mb/s · 7 Mb/s · 200 · 10 Mb/s · 50 ms · 10% | 靜止時的效能為 TCP 下載約 80 Mb/s、上傳約 7 Mb/s，UDP 接近 200 與 10 Mb/s，延遲約 50 ms，平均比有線高約 10%[53, §IV-A, Tab. II]。 |
| 證據與範圍 | 15 · 17% | 這項研究只涵蓋一個地區、一年與網路的一個世代，早於 Starlink 15 秒排程器的發現，而它的雨天數字也比之後一份雙站點預印本得到的 17% 大，所以引用時應附上地點與日期。 |
1. Where this paper sits1. 這篇論文的位置
Early Starlink studies examine complementary questions. Michel et al. measure Belgian transport, Kassem et al. examine browsing and loss, and Pan et al. map Victoria's ground segment. Ma et al. focus on weather and motion. The Ku-band user link crosses rain-affected troposphere and ionosphere, while geomagnetic storms can also disturb spacecraft operations.
These are failure modes specific to the radio and spacecraft path. The paper is also one of the few to measure a dish in a moving vehicle. This makes it the primer's main source for what Starlink does when both the terminal and the satellite are in motion.
2022 與 2023 年發表的第一波 Starlink 量測論文，各自挑了這個年輕網路的相異軸向，而這一篇挑的是地面網路研究者較少接觸的面向，也就是天氣與使用者本身的移動。Michel et al. 從比利時量測吞吐量與傳輸層行為，Kassem et al. 從歐洲與北美多個觀測點量測瀏覽與遺失，Pan et al. 從維多利亞描繪地面段，那裡與 Ma et al. 架設終端的地方只隔一段短程渡輪，而 Ma et al. 問的是使用者的體驗有多少是由終端與衛星之間的天空決定的。這個問題對第 3 章很重要，因為使用者鏈路是一條穿過對流層的 Ku 頻段射頻路徑，會被雨衰減，也是一條穿過電離層、連到位於磁層中太空載具的路徑，會被地磁風暴擾動，而這些都是射頻與太空載具路徑特有的失效模式。這篇論文也是少數量測行進車輛中終端的研究之一，使它成為本教材關於終端與衛星都在移動時 Starlink 行為的主要來源。
2. The problem, measured2. 問題，用量測說話
Stationary results show destination-dependent latency and a gap between TCP throughput and UDP capacity. Variation and loss add further costs beyond the mean.
靜止結果顯示，延遲隨目的地而變，TCP 吞吐量與 UDP 容量也有差距，變動與遺失還帶來平均值之外的成本。
| Item項目 | Values數值 | Conditions and source條件與來源 |
|---|---|---|
| Evidence and scope | Qualitative scope | The baseline the paper establishes first is what a stationary dish looks like next to cable in the same city. |
| Latency comparison | 10% · 50 ms · 1.8 ms · 8 · 23 ms · 78.4 ms | Mean latency is about 50 ms, averaging roughly 10% above cable. Destination penalties range from 1.8 ms for Singapore to 78.4 ms for Sydney. Bahrain, Tokyo, Mumbai, Cape Town, London, Northern California and São Paulo lie between 8 and 23 ms [53, §IV-A, VI-B]. |
| Dataset and workload | Qualitative scope | The more-than-fortyfold destination variation implicates routing beyond access. Later PoP studies reach a similar conclusion about geographic delay. |
| Latency comparison | 3.8 · 0.255 ms · 2.715 ms · 0.041 · 0.546 ms | Variability was the sharper difference: Starlink's latency varied about 3.8 times as much as the terrestrial link's. Its jitter was 0.255 ms down and 2.715 ms up against 0.041 and 0.546 ms for cable. |
| Capacity and rate | 66.43 Mb/s · 107.77 Mb/s · 80 Mb/s · 50.71% · 34.44% · 6.08 · 7.16 Mb/s · 193.67–205.72 Mb/s · 9.90–11.18 Mb/s | TCP download ranges from Mumbai's 66.43 to Northern California's 107.77 Mb/s, averaging about 80 Mb/s. Its variation coefficient is 50.71%, versus cable's 34.44%. TCP upload is 6.08–7.16 Mb/s, while UDP reaches 193.67–205.72 Mb/s down and 9.90–11.18 Mb/s up [53, Tab. II]. |
| Reliability and accuracy | 0.24% · 1.24% · 20% · 12 · 1.76 | At about 20% peak load, bursty loss is 0.24% down and 1.24% up, with an approximately 12-hour cycle attributed to uplink [53, §IV-B]. Second-generation dishes give 1.76× first-generation TCP upload. |
| 證據與範圍 | 定性範圍 | 論文首先建立的基準，是靜止終端在同一城市中與有線網路相比的樣子。 |
| 證據與範圍 | 10% · 50 ms · 1.8 ms · 8 · 23 ms · 78.4 ms | Starlink 的延遲平均比地面連線高約 10%，靜止時平均約 50 ms，而差距強烈取決於伺服器的位置：到新加坡為 1.8 ms，到巴林、東京、孟買、開普敦、倫敦、北加州與聖保羅介於 8 到 23 ms，到雪梨則為 78.4 ms[53, §IV-A, VI-B]。 |
| 證據與範圍 | 定性範圍 | 一個在相異目的地之間相差四十倍以上的代價，說的是存取鏈路之外的路由，，接入鏈路的影響則需分開量測，這與後來的論文把延遲追溯到接入點時得到的結論相同。 |
| 證據與範圍 | 3.8 · 0.255 ms · 2.715 ms · 0.041 · 0.546 ms | 變異性才是更明顯的差異：Starlink 延遲的變化約為地面鏈路的 3.8 倍，抖動為下行 0.255 ms、上行 2.715 ms，對照有線的 0.041 與 0.546 ms。 |
| 容量與速率 | 66.43 Mb/s · 107.77 Mb/s · 80 Mb/s · 50.71% · 34.44% · 6.08 · 7.16 Mb/s · 193.67–205.72 Mb/s · 9.90–11.18 Mb/s | 吞吐量呈現之後每項研究都確認的方向差異，TCP 下載介於到孟買的 66.43 Mb/s 與到北加州的 107.77 Mb/s 之間，平均約 80 Mb/s，變異係數 50.71%，有線為 34.44%，TCP 上傳介於 6.08 與 7.16 Mb/s，UDP 則達到下行 193.67–205.72 Mb/s、上行 9.90–11.18 Mb/s[53, Tab. II]。 |
| 證據與範圍 | 20% · 0.24% · 1.24% · 12 · 1.76 | 遺失呈突發性，在約 20% 峰值負載下為下行 0.24%、上行 1.24%，並有作者歸因於上行的約 12 小時週期[53, §IV-B]，而第二代終端的 TCP 上傳是第一代的 1.76 倍。 |
3. The design3. 設計
Rain reduces radio signal quality, the storm affects the two directions differently, and motion interrupts terminal tracking. Their measured values describe separate experimental conditions.
雨降低射頻訊號品質，風暴對上下行的影響各異，移動則會中斷終端追蹤，相關數值應依各自實驗條件解讀。
| Item項目 | Values數值 | Conditions and source條件與來源 |
|---|---|---|
| Evidence and scope | Qualitative scope | The paper uses three natural experiments as its study design, each of which stresses a separate component, as the figure shows. |
| Power and thermal conditions | Qualitative scope | Rain stresses the radio path: water droplets absorb and scatter Ku-band energy |
| Evidence and scope | Qualitative scope | which lowers the signal-to-noise ratio and forces the link onto a more robust and slower modulation |
| Capacity and rate | 27% · 215 Mb/s · 120 Mb/s · 4.1–5.2 mm · 0.34 | Precipitation reduces throughput about 27% on average. UDP download falls from about 215 to 120 Mb/s at 4.1–5.2 mm, with maximum inverse correlation 0.34. |
| Evidence and scope | 5–26% · 12 | and a separate 5–26% throughput drop above 12 °C [53, §IV-D]. |
| Capacity and rate | 3–4 · 2022 · 100 · 5 Mb/s | The 3–4 February 2022 storm reduces download from about 100 to 5 Mb/s while upload remains largely intact. This asymmetry suggests more than common atmospheric path loss. |
| Capacity and rate | 40–70 km · 80–100 Mb/s · 5–10 Mb/s | Motion stresses the terminal's ability to hold a satellite: at 40–70 km/h download stayed at about 80–100 Mb/s and upload at 5–10 Mb/s |
| Latency comparison | 100 ms · 2,800 ms | but mean latency rose to about 100 ms with a maximum of 2,800 ms |
| Evidence and scope | 16.5 s · 5 s | and outages arrived at a median interval of 16.5 s and lasted a median of 5 s |
| Evidence and scope | 95 · 162 s · 36 s | with the 95th-percentile interval at 162 s and the longest outage 36 s [53, §VI-B]. |
| Latency comparison | 10 | A stationary dish sees a satellite for about ten minutes. The authors describe gradually rising latency followed by a handover reset [53, §III-A, IV-A]. |
| 證據與範圍 | 定性範圍 | 論文以三種自然實驗作為研究設計，如下圖所示，每一個都對相異的元件施加壓力。 |
| 容量與速率 | 27% · 215 Mb/s · 4.1–5.2 mm · 120 Mb/s · 0.34 · 12 · 5–26% | 雨對射頻路徑施壓：水滴吸收並散射 Ku 頻段能量，降低訊雜比，迫使鏈路改用更強健也更慢的調變，作者量到降水期間吞吐量平均下降約 27%，UDP 下載從乾燥時的約 215 Mb/s 降到 4.1–5.2 mm 時的約 120 Mb/s，UDP 下載的最大負相關為 0.34，另外在 12 °C 以上有 5–26% 的吞吐量下降[53, §IV-D]。 |
| 容量與速率 | 2022 · 2 · 3–4 · 100 · 5 Mb/s | 2022 年 2 月 3–4 日的地磁風暴則影響另一組機制，因為下載從約 100 降到 5 Mb/s，上傳卻大致保持固定，這種方向差異需要單純路徑損耗之外的原因才能解釋，因為兩個方向穿過的是同一片大氣。 |
| 容量與速率 | 40–70 km · 80–100 Mb/s · 5–10 Mb/s · 100 ms · 2,800 ms · 16.5 s · 5 s · 95 · 162 s · 36 s | 移動對終端鎖住衛星的能力施壓：在時速 40–70 km 下，下載維持在約 80–100 Mb/s、上傳 5–10 Mb/s，但平均延遲升到約 100 ms、最大 2,800 ms，而斷線以中位數 16.5 s 的間隔出現、中位數持續 5 s，第 95 百分位間隔為 162 s，最長斷線 36 s[53, §VI-B]。 |
| 證據與範圍 | 10 | 作者也描述了靜止終端看到一顆衛星約 10 分鐘，延遲隨衛星遠離而逐漸上升，直到換手將其重設[53, §III-A, IV-A]。 |
Original figure note原始圖說出處
Drawn for this primer from Ma et al., IEEE INFOCOM 2023, Secs. IV-D and VI-B.4. Evaluation setup4. 實驗設定
The study uses live dishes, weather records and vehicle measurements. It characterizes access paths and terrestrial routing under the network conditions of its observation period.
研究使用實際終端、天氣紀錄與車輛量測，描述觀測期間網路條件下的接入路徑及地面路由。
| Item項目 | Values數值 | Conditions and source條件與來源 |
|---|---|---|
| Evidence and scope | Qualitative scope | Vancouver end users test first- and second-generation dishes against a cable control. Cloud targets are Bahrain, Cape Town, London, Mumbai, Northern California, São Paulo, Singapore, Sydney and Tokyo [53, §III–IV, Tab. II]. |
| Dataset and workload | 40–70 km | Weather data were matched to throughput samples to obtain the rain and temperature correlations, the storm effect is a before-and-during comparison around one event. The mobility results come from driving at 40–70 km/h [53, §IV-D, VI-B]. |
| Evidence and scope | 2022 | The study leaves ISL use unspecified. A 2022 Vancouver terminal would normally use a nearby gateway, so results concern access plus onward terrestrial routing. |
| Evidence and scope | Qualitative scope | The calc table turns four of the reported figures into the ratios the primer uses. |
| 證據與範圍 | 定性範圍 | 實驗是實測，範圍限定為實際接入路徑的量測，由溫哥華地區的終端使用者以第一代與第二代 Starlink 終端執行，並以地面有線連線作為對照，對象是位於巴林、開普敦、倫敦、孟買、北加州、聖保羅、新加坡、雪梨與東京的雲端伺服器[53, §III–IV, Tab. II]。 |
| 軌道配置 | 40–70 km | 天氣資料與吞吐量樣本配對以得到雨量與溫度的相關性，風暴效應是圍繞單一事件的前後比較，而移動結果來自時速 40–70 km 的行車[53, §IV-D, VI-B]。 |
| 證據與範圍 | 2022 | 流量的星間路徑仍待確定，而在 2022 年，溫哥華地區的終端通常會經由彎管連到附近的閘道，所以結果描述的是存取鏈路及其後的地面路由。 |
| 證據與範圍 | 定性範圍 | 下方計算表把四個報告數字轉成本教材使用的比例。 |
| Question問題 | Calculation算式 | Result結果 |
|---|---|---|
| Drop in UDP download at 4.1–5.2 mm of rain4.1–5.2 mm 雨量時 UDP 下載的降幅 | \(1 - 120/215\) | ≈44%computed. Inputs from Ma et al., Sec. IV-D. The 27% headline is the average over all rain計算值。輸入來自 Ma et al., Sec. IV-D。27% 的標題數字是所有降雨的平均 |
| Download remaining during the storm風暴期間剩餘的下載 | \(5/100\) | 5%computed. Sec. IV-D, one event計算值。Sec. IV-D，單一事件 |
| Download-to-upload ratio, TCP and UDP下載對上傳比，TCP 與 UDP | \(80/7,\ 200/10\) | ≈11 : 1, 20 : 1computed. Averages from Table II計算值。平均值來自 Table II |
| Outage share if each median 16.5 s gap held one median 5 s outage若每個中位數 16.5 s 間隔含一次中位數 5 s 斷線的斷線比例 | \(5/16.5\) | ≈30%computed. An upper-end reading, since combining distribution medians gives an approximation計算值。屬偏高的讀法，因為組合兩個分布中位數所得的是近似值 |
5. What was measured5. 量到了什麼
| Claim宣稱 | Setup and caveat設定與附註 |
|---|---|
| Precipitation lowers throughput about 27% on average. UDP download ≈215 → ≈120 Mb/s at 4.1–5.2 mm [53, §IV-D]降水使吞吐量平均降低約 27%。UDP 下載在 4.1–5.2 mm 時 ≈215 → ≈120 Mb/s[53, §IV-D] | One region. Correlation at most 0.34. WetLinks found 17% at two European sites, which is a preprint. preprint單一地區。相關性最高 0.34。WetLinks 在兩個歐洲站點得到 17%，屬預印本。preprint |
| Geomagnetic storm of 3–4 Feb 2022: download ≈100 → 5 Mb/s, upload largely unaffected [53, §IV-D]2022 年 2 月 3–4 日地磁風暴：下載 ≈100 → 5 Mb/s，上傳大致保持固定[53, §IV-D] | A single event. Richter et al. later found latency appears stable across storm intensities, a separate metric and so a separate observation.單一事件。Richter et al. 之後發現各種風暴強度下的延遲似乎穩定，這項觀察使用相異指標。 |
| Stationary TCP ≈80 / 7 Mb/s, UDP ≈200 / 10 Mb/s. Gen2 TCP upload 1.76× Gen1 [53, Tab. II]靜止 TCP ≈80 / 7 Mb/s，UDP ≈200 / 10 Mb/s。第二代 TCP 上傳為第一代的 1.76 倍[53, Tab. II] | 2022 network. The TCP-to-UDP gap suggests the transport, the transport as well as the link, limited TCP.2022 年的網路。TCP 與 UDP 的差距顯示TCP 的限制還包括傳輸層。 |
| Latency ≈50 ms stationary, ≈10% above cable. Gap 1.8 ms (Singapore) to 78.4 ms (Sydney). Variation ≈3.8× cable [53, §IV-A, VI-B]靜止延遲 ≈50 ms，比有線高 ≈10%。差距由 1.8 ms（新加坡）到 78.4 ms（雪梨）。變異 ≈ 有線的 3.8 倍[53, §IV-A, VI-B] | The destination-dependent gap points to routing beyond the PoP, which the paper leaves combined.隨目的地而異的差距指向 PoP 之外的路由，各項貢獻仍待分解。 |
| In motion at 40–70 km/h: outages at a median 16.5 s interval lasting a median 5 s. Mean RTT ≈100 ms, max 2,800 ms [53, §VI-B]時速 40–70 km 行進中：斷線中位數間隔 16.5 s、中位數持續 5 s。平均 RTT ≈100 ms，最大 2,800 ms[53, §VI-B] | Dish type, mounting and road environment matter and await systematic variation.終端類型、安裝方式與道路環境都有影響，其系統性變化測試仍待完成。 |
| Bursty loss 0.24% down / 1.24% up at ≈20% of peak, with a ≈12-hour cycle attributed to the uplink [53, §IV-B]突發性遺失在 ≈20% 峰值時為下行 0.24% / 上行 1.24%，並有歸因於上行的 ≈12 小時週期[53, §IV-B] | Light load. Michel et al. saw 1.56–1.96% under load in Belgium.輕負載。Michel et al. 在比利時於負載下看到 1.56–1.96%。 |
6. Limits and open questions6. 範圍與待解問題
- One region, one season, one network generation. All dishes were in the Vancouver area in 2022. Thus, the rain figure reflects that climate and the 2022 constellation. The later two-site WetLinks preprint measured a smaller 17% median drop in rain windows. Rain-fade values in decibels from ITU-R P.618 remain to be retrieved for the primer. Thus, the measured throughput effects are the only rain numbers it quotes.一個地區、一個季節、一個網路世代。所有終端都在 2022 年的溫哥華地區，所以雨天數字反映的是那裡的氣候與 2022 年的星座，而之後的雙站點 WetLinks 預印本在雨天時段量到較小的 17% 中位數降幅。本教材的待補充資料包括 ITU-R P.618 以分貝表示的雨衰數值，所以它引用的雨天數字只有實測的吞吐量效應。
- The storm is one event with an unresolved mechanism. Download collapse with preserved upload suggests direction-specific effects. Identifying scintillation, satellite operations or gateway behavior needs more evidence.風暴觀測來自單一事件，機制仍待確認。下載崩落而上傳完好十分引人注目，但單一的前後比較要辨明原因仍需額外證據，候選機制包括電離層閃爍、風暴期間的衛星操作還是閘道效應，而這種方向差異是論文留給後續研究的線索。
- Visibility and scheduled reconfiguration have separate timescales. The described ten-minute satellite pass concerns visibility. Later work finds global reassignment at :12/:27/:42/:57. Mohan et al. observe those boundaries with one satellite visible, supporting a scheduling explanation for stationary latency steps.後續研究區分了衛星可見性與排程重組。論文描述一顆衛星服務一支終端約 10 分鐘、延遲上升直到換手，這是一個可見性的陳述。Tanveer et al. 之後顯示 Starlink 依全域 15 秒排程在每分鐘第 12、27、42、57 秒重新指派使用者，而 Mohan et al. 在只有一顆衛星可見時也看到相同的邊界，所以靜止時的延遲階躍現在應被解讀為排程式重新配置，，十分鐘過境則描述可見性。
- Moving outages need timing attribution. Their 16.5 s median interval resembles the later-discovered 15-second schedule. Alignment with reconfiguration, obstruction or both remains an open measurement.16.5 秒的斷線間隔是排程器嗎？移動中斷線之間的中位數間隔接近 15 秒的重新配置間隔，但論文早於該發現，對齊檢驗則待完成，所以行進中的終端是在重新配置邊界、遮蔽事件還是兩者都會失去服務，仍是一個待量測的問題。
- Transport diagnosis remains open. TCP reached about 80 Mb/s while UDP reached about 200 Mb/s. The paper reports the gap while leaving the respective effects of congestion control and loss open. Thus, the attribution of the shortfall belongs to the transport papers, notably Kassem et al.'s BBR result and Garcia et al.'s drop-front queue.傳輸層診斷仍待完成。TCP 約 80 Mb/s 而 UDP 約 200 Mb/s，論文報告了差距，壅塞控制行為與遺失的各自貢獻則待分解，所以這個落差的歸因屬於傳輸層論文，特定是 Kassem et al. 的 BBR 結果與 Garcia et al. 的前端丟棄佇列。
7. How it connects7. 與其他內容的關聯
- Section 3.4 compares rain, storm and 80/7 Mb/s measurements with WetLinks' 17% throughput decline and Kassem's doubled page transit time.第 3.4 節：雨、風暴與 80/7 Mb/s 的數字列在該章的實測效能表中，旁邊是 WetLinks 的 17% 與 Kassem et al. 加倍的網頁傳輸時間。
- Section 3.5: the ten-minute dwell and the 16.5 s mobile outages are reread there against the 15-second schedule and the handover studies.第 3.5 節：十分鐘駐留與 16.5 s 的行動斷線在那裡以 15 秒排程與換手研究重新解讀。
- Section 3.1: rain fade acts on the Ku waveform that Humphreys et al. reverse-engineered, by forcing the link toward its more robust modulation.第 3.1 節：雨衰作用在 Humphreys et al. 逆向解析的 Ku 波形上，迫使鏈路轉向較強健的調變。
- Related studies: pimrc23-pan maps the nearby ground segment, imc22-kassem examines rain and browsing, conext23-tanveer explains scheduling, and taes23-humphreys describes the affected waveform.相關導讀：pimrc23-pan 從同一地區的維多利亞描繪地面段，imc22-kassem 量測雨對瀏覽的影響，conext23-tanveer 補上本文範圍之外的排程器，taes23-humphreys 則描述了被天氣劣化的波形。
Key numbers關鍵數字
| Number | 數字 | Meaning (setup) | 意義（設定） |
|---|---|---|---|
| −27% | Average throughput drop during precipitation, Vancouver area 2022 [53, §IV-D] | 降水期間平均吞吐量降幅，2022 年溫哥華地區[53, §IV-D] | |
| 215 → 120 Mb/s | UDP download, dry versus 4.1–5.2 mm rain [53, §IV-D] | UDP 下載，乾燥對 4.1–5.2 mm 雨量[53, §IV-D] | |
| 100 → 5 Mb/s | Download during the 3–4 Feb 2022 geomagnetic storm. Upload largely unaffected [53, §IV-D] | 2022 年 2 月 3–4 日地磁風暴期間的下載。上傳大致保持固定[53, §IV-D] | |
| ≈80 / 7 Mb/s | Stationary TCP download / upload averages. UDP ≈200 / 10 [53, Tab. II] | 靜止 TCP 下載 / 上傳平均。UDP ≈200 / 10[53, Tab. II] | |
| ≈50 ms · +10% | Stationary latency · average penalty versus cable [53, §IV-A, VI-B] | 靜止延遲 · 相對有線的平均代價[53, §IV-A, VI-B] | |
| 16.5 s · 5 s | Median interval between outages · median outage length at 40–70 km/h [53, §VI-B] | 時速 40–70 km 時斷線的中位數間隔 · 中位數斷線長度[53, §VI-B] | |
| ≈100 · 2,800 ms | Mean · maximum RTT while driving [53, §VI-B] | 行車時的平均 · 最大 RTT[53, §VI-B] | |
| 0.24% / 1.24% | Bursty loss down / up at ≈20% of peak load [53, §IV-B] | ≈20% 峰值負載下的突發遺失，下行 / 上行[53, §IV-B] |