Digest · Topology and routing · read after Chapter 5
導讀 · 拓樸與路由 · 第 5 章後閱讀
Network Topology Design at 27,000 km/hour時速 27,000 公里下的網路拓樸設計
A motif repeats a local link pattern between co-moving satellites, preserving its links through an orbit. The study scores patterns by stretch and hops. Longer shortcuts improve this objective, while shorter range limits the gain. The results provide a dated design baseline.motif 在同向運動的衛星間重複局部連線規則，使鏈路維持整圈軌道。研究以拉伸比與跳數評分，較長的捷徑改善目標值，縮短射程則限制增益。結果提供一套附日期的設計基準。
If you read nothing else
只讀這一段也可以
Bhattacherjee and Singla optimize four laser links per satellite in a constellation moving at 27,000 km/h. They seek short distances and few hops. A motif assigns two same-direction neighbors to every satellite through an identical local rule. Repeating it preserves link feasibility throughout the orbit. Each laser retains its acquired partner. [1, §4.1–4.2]Bhattacherjee 與 Singla 問的是，一個以時速 27,000 公里運動的星座應如何為每顆衛星的四條雷射鏈路接線，使路徑在公里數與跳數上都短，而他們的答案是 motif，也就是每顆衛星向兩顆同向運動的衛星各接一條鏈路的局部模式，在整個星座中以相同方式重複，使每條鏈路在整個軌道週期內都保持可行，各具雷射皆能維持已擷取的對象[1, §4.1–4.2]。
| Measure or comparison參數或比較項目 | Values, conditions and evidence數值、條件與證據 |
|---|---|
| Path length and hops | The population-product traffic matrix covers the 1,000 largest cities. On a 40×40 reference, one motif reduces traffic-weighted stretch-plus-hop objective Φ1 by 44% versus +Grid. A three-zone motif changes links about every 12 minutes. For 24×66 Starlink, gains are 54% at 5,014 km range and 40% at the 2,006 km +Grid minimum. Kuiper gains are 45% at 5,440 km and 4% at 1,761 km. [1, §4.4] [1, §5.4, §6, Tab. 1] |
| Path length and hops | On the 40×40 reference, +Grid averages 1.25 stretch and 10.57 hops. Results assume the November 2018 24×66 layout and four terminals, both superseded later. [1, §4.3] |
| 路徑長度與跳數 | 以「拉伸加跳數」的合併目標在 1,000 個人口最多城市之間的人口乘積流量上計分，單一 motif 在 40×40 參考星座上就已把流量加權目標 Φ1 比 +Grid 降低 44%[1, §4.4]，而每顆衛星約每 12 分鐘更換一次鏈路的三區多 motif，把改善提高到 24×66 Starlink 殼層在 5,014 公里可見性限制範圍下的 54%、在 +Grid 本身所需的 2,006 公里最小範圍下的 40%、Kuiper 在 5,440 公里下的 45%，而 Kuiper 在 1,761 公里下只有 4%[1, §5.4, §6, Tab. 1]。 |
| 路徑長度與跳數 | 40×40 星座上的 +Grid 基準平均拉伸為 1.25、跳數 10.57[1, §4.3]，而所有結果都建立在 2018 年 11 月的 24×66 佈局與每顆衛星四個終端之上，兩者後來都已被取代。 |
1. Where this paper sits1. 這篇論文的位置
This is the first full-length paper on inter-satellite topology design and the quantitative baseline that every later wiring proposal measures itself against. Thus, Chapter 5 gives it the center of Section 5.2. The same group's HotNets 2018 agenda used +Grid held static for one minute. This paper develops that work before Hypatia adopts +Grid as its default.
The enduring result is the stable-topology argument. Links changing every few minutes incur setup times of tens of seconds. The paper therefore favors patterns stable by construction, alongside the 54% gain. Ali and Zhang (2026) find motifs depend on complete symmetric deployment.
TinyLEO instead uses a topology-recompiling controller that favors long-lived links. Both retain the emphasis on link stability.
這是第一篇關於星間拓樸設計的完整論文，也是後來每個接線提案都以之為對照的量化基準，所以第 5 章把它放在第 5.2 節的中心。它直接承接同一群人 HotNets 2018 的議程論文，該論文假設 +Grid 並每次保持一分鐘靜態，而它又先於把 +Grid 作為預設寫進模擬器的 Hypatia。這篇論文持久的貢獻在於穩定拓樸的設計框架，54% 則是特定設定下的結果：它是「每隔幾分鐘就改變的鏈路建鏈成本偏高，因為雷射建鏈要花數十秒，所以拓樸應該在結構上本就穩定的模式中選擇」這一論證的標準參考。後來的研究繼承這個框架但對細節提出異議，Ali 與 Zhang 在 2026 年發現 motif 需要完整對稱部署，真實殼層則為部分部署，因為它假設完整且對稱的部署，而 TinyLEO 以一個在重新編譯拓樸時偏好長壽鏈路的控制器取代固定模式。
2. The problem, measured2. 問題，用量測說話
The authors show this concretely with the two standard tools. Acquisition takes a few seconds to tens of seconds. Rapidly changing topologies spend much of their operation reacquiring partners. The calculation table restates the objective and +Grid baseline. [1, §2.3]作者用兩種標準工具具體說明這一點。由於一次鏈路變更需要數秒到數十秒[1, §2.3]，在這個時間尺度上變動的拓樸會把大部分生命花在重新對準。計算表重述目標函數與作為改善對照的 +Grid 基準。
| Measure or comparison參數或比較項目 | Values, conditions and evidence數值、條件與證據 |
|---|---|
| Evaluation conditions | Traffic-matrix graph optimization becomes difficult at 27,306 km/h. At 550 km altitude, satellites cover roughly Munich-to-Berlin distance each minute. [1, §2.2] |
| Evaluation conditions | An ILP for 20 cities lowers Φ1 by 54% versus +Grid. At 25 cities, it exceeds two days on a 64-core machine. The authors extrapolate 10²⁹ days for 1,000 cities. Solutions one minute apart share only 9% of links. [1, §3.1] |
| Path length and hops | Feasible random regular graphs cut median hops by 53% while increasing stretch by 11%. Different partner directions make over 8% of links infeasible within two minutes. After five minutes, 19% are infeasible. [1, §3.2, Fig. 5] |
| Power budget | At 550 km, clearing 80 km of atmosphere permits 5,014 km links. The shortest connected +Grid ranges are 2,006 km for Starlink and 1,761 km for Kuiper. These represent the power-limited worst cases. [1, §2.3, Fig. 2] [1, §6] |
| Plane configuration | For a 53°-inclined 40×40 constellation, candidates rise from about 1,100 at the Equator to 3,600 at 53°. A polar constellation rises from about 600 to 4,000 at 90°. [1, §5.1, Fig. 12] |
| 軌道高度 | 問題在於，設計網路拓樸的常規方法，也就是為一個流量矩陣挑選最佳的圖，在節點以時速 27,306 公里移動時就失效了，這在 550 公里高度相當於每分鐘跨越大約慕尼黑到柏林的距離[1, §2.2]。 |
| 評估條件 | 以 20 個城市求解的整數線性規劃（ILP）讓 Φ1 比 +Grid 低 54%，但 25 個城市在 64 核心的機器上兩天內就仍待求解完成，作者推估 1,000 個城市需要 10²⁹ 天，而且相隔一分鐘算出的兩個 ILP 拓樸只共用 9% 的鏈路[1, §3.1]。 |
| 路徑長度與跳數 | 限制在可行鏈路上的隨機正則圖把中位跳數降低 53%，代價是拉伸增加 11%，但兩分鐘內就有超過 8% 的鏈路、五分鐘內有 19% 的鏈路變得超出可行條件，因為它們連接的衛星各有運動方向[1, §3.2, Fig. 5]。 |
| 功率預算 | 星座再透過鏈路範圍限制搜尋，從 550 公里出發、最低點必須高於 80 公里大氣層的鏈路最遠可達 5,014 公里[1, §2.3, Fig. 2]，而 +Grid 仍能連通的最短範圍，Starlink 為 2,006 公里、Kuiper 為 1,761 公里，被當作功率受限雷射的最壞情況[1, §6]。 |
| 軌道面配置 | 在可見性限制內，可能的 motif 數量隨著軌道面匯聚而隨緯度增加，53° 傾角的 40×40 星座從赤道約 1,100 個增加到 53° 約 3,600 個，極軌道星座則從約 600 個增加到 90° 約 4,000 個[1, §5.1, Fig. 12]。 |
| Question | 問題 | Formula | 公式 | Result | 結果 |
|---|---|---|---|---|---|
| Stretch of a satellite path between two cities兩城市之間衛星路徑的拉伸 | \( S=\tfrac{L_{sat}}{L_{geodesic}} \) | +Grid on the 40×40 constellation: 1.25definition [1, §3.1]. Population-weighted average at one snapshot [1, §4.3]定義[1, §3.1]。單一快照的人口加權平均[1, §4.3] | |||
| Combined objective per city pair, and its traffic-weighted sum每對城市的合併目標及其流量加權總和 | \( M_\alpha=\alpha S+B,\quad \Phi_\alpha=\textstyle\sum_{s,t} H_{st}\,M_\alpha^{st} \) | +Grid on 40×40: 10.57 hops. α = 1, 5, 10 triedB is the ISL hop count and H the population-product traffic matrix [1, §3.1, §5.5]. Table 1 reports the median of M1 across city pairs and Φ1B 為 ISL 跳數，H 為人口乘積流量矩陣[1, §3.1, §5.5]。表 1 報告 M1 在城市對上的中位數與 Φ1 | |||
| One-way propagation over the longest and shortest Starlink link ranges considered所考慮的最長與最短 Starlink 鏈路範圍的單向傳播延遲 | \( t=\tfrac{d}{c} \) | 5,014 km → 16.7 ms. 2,006 km → 6.69 mscomputed. Ranges [1, §2.3, §6], c = 299,792,458 m/s. 5,014 km is the longest link from 550 km whose lowest point clears 80 km計算值。範圍[1, §2.3, §6]，c = 299,792,458 m/s。5,014 公里是從 550 公里出發、最低點高於 80 公里的最長鏈路 |
3. The design3. 設計
Every satellite applies the same two-link rule. Candidate partners stay within range and travel in the same direction. Thus, a motif is a three-satellite, two-link pattern whose repetition fully specifies the topology [1, §4.1, Fig. 6] . Each satellite initiates two links and receives two more, using the four ISLs in Starlink's revised filings. Partners move together, retaining links for the orbit. Relative velocities rise only at high latitudes, as with +Grid. [1, §3.1, fn. 3] [1, §4.2]motif 是每顆衛星都以相同方式套用的規則，從射程內且同向運動的衛星中選兩顆相連，所以一個 motif 就是一個「三顆衛星、兩條鏈路」的模式，把它重複到整個星座便完整決定了拓樸[1, §4.1, Fig. 6]。因為每顆衛星主動建立兩條鏈路，又從規則指向它的衛星接收兩條，所以每顆衛星用上 Starlink 修訂後申請文件所示的四條 ISL[1, §3.1, fn. 3]，而由於夥伴與它一起運動，鏈路在整個軌道週期內都保持，相對速度只在高緯度增大，與 +Grid 鏈路完全相同[1, §4.2]。+Grid 本身就是一個 motif，如圖 A 左側所繪。由於高緯度的衛星彼此較近，論文接著讓 motif 隨緯度變化，以寬度 W 的帶狀區從赤道向外逐區決定 motif。
| Measure or comparison參數或比較項目 | Values, conditions and evidence數值、條件與證據 |
|---|---|
| Motif search | The +Grid is one motif, shown at left in Figure A. Search enumerates candidate neighbor pairs at the Equator, where separation is largest. This ensures orbital feasibility, and symmetry removes equivalent patterns. This leaves 1,029 for the 40×40 constellation, and scores each by shortest-path stretch and hop count under the population-product traffic matrix [1, §4.1, §4.3] . |
| Plane configuration | The center of Figure A shows links skipping planes and indices. Longer individual links reduce total path hops. The paper's Pareto plot shows that a motif giving up 2% of stretch cuts hops by 32% [1, §4.3, Fig. 7] . |
| Constellation configuration | Closer high-latitude spacing permits latitude-dependent motifs. Zones extend outward from the Equator in bands of width W, with W = 18°. Three zones cover latitude up to 53°. Each satellite changes ISLs about every 12 minutes. The authors judge setup of a few seconds to small tens of seconds tolerable. Gains beyond a single motif saturate near 7%. [1, §5.3–5.4, Fig. 13] |
| Throughput | The abstract's "nearly 2× the throughput" rests on hop count as the throughput proxy. This occurs because a connection crossing fewer ISLs consumes capacity on fewer links. A shortest-path congestion check supports the proxy. The +Grid 75th- and 90th-percentile link loads are 4× and 5× the Φ1-optimized multi-motif loads. [1, §2.4, §5.6, Fig. 15] |
| 星座配置 | 搜尋時列舉赤道上一顆衛星射程內的所有衛星對，因為赤道處衛星彼此最遠，所以每條候選鏈路在任何位置都可行，再刪去因對稱而等價的 motif，40×40 星座剩下 1,029 個，並以人口乘積流量矩陣下的最短路徑拉伸與跳數為每個 motif 計分[1, §4.1, §4.3]。 |
| 路徑長度與跳數 | 圖 A 中間顯示搜尋會找到的那類替代方案，鏈路跳過軌道面與編號，使橫越星座的路徑需要較少跳數，代價是單條鏈路較長，而論文的 Pareto 圖顯示，犧牲 2% 拉伸的 motif 可以減少 32% 的跳數[1, §4.3, Fig. 7]。 |
| 換鏈頻率 | W = 18° 時到 53° 共有三區，衛星約每 12 分鐘更換一次 ISL，作者認為這個頻率使數秒到十幾秒的建鏈開銷可以容忍，而相對單一 motif 的增益在約 7% 處飽和[1, §5.3–5.4, Fig. 13]。 |
| 吞吐量 | 摘要中「近 2 倍的吞吐量」是以跳數作為吞吐量的代理，因為經過較少 ISL 的連線只佔用較少鏈路的容量，作者並以最短路徑路由下的壅塞檢驗支持這個代理，其中 +Grid 第 75 與第 90 百分位的鏈路負載各自是 Φ1 最佳化多 motif 的 4 倍與 5 倍[1, §2.4, §5.6, Fig. 15]。 |
4. Evaluation setup4. 實驗設定
Evaluation uses Python and networkx shortest paths. Emulation and measurement lie outside its scope. Code and data are public. [1, §2.4, §5.2] [1, §1]評估是以 Python 撰寫、建立在 networkx 最短路徑演算法上的模擬[1, §2.4, §5.2]，模擬平台及量測皆位於範圍之外，程式碼與資料都已公開[1, §1]。延遲只計傳播，並以跳數代表吞吐量，所以鏈路速率、排隊及節點延遲皆位於模型範圍之外[1, §2.4]。
| Measure or comparison參數或比較項目 | Values, conditions and evidence數值、條件與證據 |
|---|---|
| Constellation configuration | The constellations are Starlink's first phase as filed by July 2019, 1,584 satellites in 24 planes of 66 at 550 km and 53° |
| Kuiper configuration | Kuiper's phase A, a 34×34 constellation of 1,156 satellites at 630 km and 51.9° |
| Reference constellation | and a uniform 40×40 constellation at 53° and 550 km used as the reference throughout Sections 3 to 5 |
| ISL count | with four ISLs on every satellite [1, §2.1, §3.1, §6]. |
| Evaluation conditions | The link range runs between the visibility limit, 5,014 km for Starlink and 5,440 km for Kuiper. The +Grid minimum is 2,006 km for Starlink and 1,761 km for Kuiper. The 40×40 minimum is its longest +Grid link, 1,467 km. Its Φ1 approaches the best result by 3,000 km. [1, §4.6, §6, Fig. 11] |
| PoP coverage | Traffic uses population products across the 1,000 largest cities, with 2025 population estimates. Each city has one unlimited-bandwidth station reaching any in-range satellite. A GDP-product check uses the top 100 cities. The 40×40 motif gain then drops from 48% to 34%. [1, §3.1, §4.7] |
| Latency comparison | Latency includes propagation only, with zero node delay. Link rate and queueing lie outside the model, while hops proxy throughput. Minute-by-minute two-hour runs supplement Section 4.3 snapshots. The best single motif's Φ1 stays within 10% of its median. [1, §2.4] [1, §4.4, Fig. 8] |
| 星座配置 | 星座為截至 2019 年 7 月申請內容的 Starlink 第一階段，1,584 顆衛星、24 個軌道面各 66 顆、550 公里、53°。 |
| 評估條件 | Kuiper 的 A 階段，即 630 公里、51.9° 的 34×34 星座，共 1,156 顆。 |
| 星座配置 | 以及 53°、550 公里的均勻 40×40 星座，作為第 3 到第 5 節全程使用的參考，且每顆衛星都有四條 ISL[1, §2.1, §3.1, §6]。 |
| 評估條件 | 鏈路範圍介於可見性極限（Starlink 5,014 公里、Kuiper 5,440 公里）與 +Grid 所需的最小值（2,006 公里與 1,761 公里）之間，而 40×40 星座的下限是其最長的 +Grid 鏈路 1,467 公里，且在 3,000 公里時 Φ1 已接近最佳[1, §4.6, §6, Fig. 11]。 |
| 流量矩陣與敏感度 | 流量是以 2025 年人口估計、1,000 個人口最多城市的人口乘積矩陣，每個城市只有一座頻寬視為充足、可連到射程內任何衛星的地面站，另以前 100 大城市的 GDP 乘積矩陣作為檢驗，在其上 40×40 的 motif 增益從 48% 降為 34%[1, §3.1, §4.7]。 |
| 評估條件 | 第 4.3 節的單一快照結果另有兩小時逐分鐘的模擬補充，其中最佳單一 motif 的 Φ1 在中位數上下變動至多為 10%[1, §4.4, Fig. 8]。 |
5. What was measured5. 量到了什麼
| Claim宣稱 | Setup and caveat設定與附註 |
|---|---|
| Starlink 24×66 with three zones gains 52% median M1 and 54% Φ1 at 5,014 km. At 2,006 km, gains are 37% and 40%. [1, §6, Tab. 1]Starlink 24×66、三區多 motif：在 5,014 公里下相對 +Grid 的中位 M1 改善 52%、Φ1 改善 54%，在 2,006 公里下為 37% 與 40%[1, §6, Tab. 1]。 | Population-product traffic over 1,000 cities. The abstract's "54%" is the Φ1 figure. Starlink's few planes contain many satellites. This raises the +Grid minimum range and supplies more candidates for longer links, explaining short-range robustness.1,000 個城市的人口乘積流量，摘要的「54%」是 Φ1 數字。作者把 Starlink 在短範圍下的穩健歸因於軌道面少而每面衛星多，這既拉高了 +Grid 的最小範圍，也提供更多較長鏈路的候選。 |
| Kuiper 34×34 gains 38% / 45% at 5,440 km and 1% / 4% at 1,761 km. The 40×40 reference gains 45% / 48% at 5,014 km. At minimum range, it gains 9% / 7%. [1, §6, Tab. 1]Kuiper 34×34：在 5,440 公里下 38%／45%，在 1,761 公里下降到 1%／4%。40×40 參考：在 5,014 公里下 45%／48%，在其最小範圍下 9%／7%[1, §6, Tab. 1]。 | Table 1 orders constellations as Starlink, Kuiper and 40×40, listing median M1 before Φ1. The 45% / 48% / 54% slide values mix these entries. Kuiper Φ1 and 40×40 median M1 both gain 45%. The 40×40 Φ1 gain is 48%, and Starlink Φ1 gains 54%. The introduction also gives the Starlink and Kuiper values. [1, §1]表 1 的列序為 Starlink、Kuiper、40×40，每列先中位 M1 再 Φ1，這也釐清了演講投影片中 45%／48%／54% 的順序：45% 是 Kuiper 的 Φ1 改善（也是 40×40 的中位 M1 改善），48% 是 40×40 的 Φ1 改善，54% 是 Starlink 的，引言對 Starlink 與 Kuiper 的敘述也相同[1, §1]。 |
| +Grid baseline on the 40×40 constellation: population-weighted average stretch 1.25 and 10.57 hops [1, §4.3].40×40 星座上的 +Grid 基準：人口加權平均拉伸 1.25、跳數 10.57[1, §4.3]。 | One arbitrarily chosen snapshot. The paper leaves absolute +Grid baselines unreported for Starlink or Kuiper, only the percentages of Table 1.單一任意選取的快照。論文Starlink 與 Kuiper 的 +Grid 絕對基準皆仍待補充，只有表 1 的百分比。 |
| Single motif on 40×40: median (95th-percentile) M1 improves by 44.5% (54%). A motif giving up 2% (10%) of stretch cuts hops by 32% (47%) [1, §4.3, Fig. 7].40×40 上的單一 motif：中位（第 95 百分位）M1 改善 44.5%（54%），犧牲 2%（10%）拉伸的 motif 可減少 32%（47%）跳數[1, §4.3, Fig. 7]。 | One snapshot. Over two hours of minute-by-minute snapshots the best motif keeps Φ1 44% below the +Grid in the median [1, §4.4].單一快照。在兩小時逐分鐘的快照中，最佳 motif 的 Φ1 中位數仍比 +Grid 低 44%[1, §4.4]。 |
| "Nearly 2× the throughput of standard past methods" and a "∼2× efficiency improvement over the neighbor-grid baseline" from a single motif [1, Abstract, §1].「標準既有方法近 2 倍的吞吐量」以及單一 motif「相對鄰居格狀基準約 2 倍的效率改善」[1, Abstract, §1]。 | Throughput is proxied by hop count. Thus, the claim is a topology result rather than a transport measurement. Link rate and absolute capacity remain unspecified, and the congestion check of Section 5.6 is the only load-based support.吞吐量以跳數代理，因此這是拓樸結果傳輸量測則須另行確認。論文鏈路速率與絕對容量皆仍待補充，第 5.6 節的壅塞檢驗是唯一以負載為基礎的佐證。 |
| ILP solutions one minute apart share 9% of links. Random graphs lose over 8% of feasible links within 2 minutes and 19% within 5 minutes. [1, §3.1–3.2]鏈路動態：相隔一分鐘算出的 ILP 拓樸只共用 9% 的鏈路，隨機圖的鏈路有超過 8%（19%）在 2（5）分鐘內變得超出可行條件[1, §3.1–3.2]。 | For the 20-city ILP and for random graphs, used as motivation. Single-motif topologies have zero churn by construction [1, §4.2].針對 20 個城市的 ILP 與隨機圖，作為動機。單一 motif 拓樸在結構上零變動[1, §4.2]。 |
| A later preprint reuses motifs on a synthetic full four-ISL Starlink shell. Balanced motifs give 48.9 ms and 21.5 hops. +Grid gives 60.6 ms and 23.5 hops. [34, Tab. 2]被重用為基準：在四條 ISL 的合成完整 Starlink 殼層上，平衡 motif 給出 48.9 ms 與 21.5 跳，對比 +Grid 的 60.6 ms 與 23.5 跳[34, Tab. 2]preprint。 | That reimplementation uses 72×22, compared with the original 24×66. Its roughly 9% hop saving is below CoNEXT's 32–53% reductions on 40×40. [1, §4.3, §5.5] The same authors report that motifs require full deployment for direct application.在 72×22 佈局上，對照原研究的 24×66，且 motif 由預印本作者重新實作，其約 9% 的跳數節省遠低於 CoNEXT 在 40×40 上報告的 32–53% 跳數減少[1, §4.3, §5.5]。同一作者也報告 motif 直接套用時須滿足完整部署條件。 |
6. Limits and open questions6. 限制與待解的問題
- The ground segment and traffic are idealized. Each city has one unlimited-bandwidth station reaching any in-range satellite. Joint ground-station and ISL design remains future work. The traffic matrix is static, with time-varying traffic also deferred. Thus, the gains are those of the ISL graph alone [1, §3.1, §4.7, §7] .地面段與流量都被理想化。每個城市只有一座頻寬視為充足、可連到射程內任何衛星的地面站，地面站與 ISL 的聯合設計留待後續，流量矩陣也是靜態的，隨時間變化的流量同樣延後處理，所以這些增益只屬於 ISL 圖本身[1, §3.1, §4.7, §7]。
- The constellation and terminal count are superseded. The August 2019 change from 24×66 to 72×22 increases intra-plane separation from 659 km to 1,969 km. This changes range-feasible motifs. The four terminals have become three with an undisclosed topology (fact base Part C, items 12, 13 and 55). Ali and Zhang found that a three-ISL grid is disconnected on real ephemerides. Thus, the motif search would have to be redone under the revised degree constraint.星座與終端數已被取代。24×66 殼層在 2019 年 8 月變為 72×22，面內間距從 659 公里變為 1,969 公里，也改變了哪些 motif 符合範圍限制，而四個終端已變成三個且拓樸仍待公開（事實庫 C 部第 12、13 與 55 項）。Ali 與 Zhang 發現三 ISL 網格在真實星曆上是分成多個連通分量的，所以 motif 搜尋必須在相異的度數限制下重做。
- Full symmetric deployment is assumed. A motif rule presumes every slot in every plane is occupied. Ali and Zhang find motifs unsuitable for partially filled real Shell 1. Their stable-link designs on 90 days of TLEs achieve 52.9 ms and 8.2 hops. Adapted +Grid gives 89.0 ms and 23.4 hops. The motif result therefore bounds regular-wiring gains. preprint假設完整對稱的部署。motif 規則預設每個軌道面的每個槽位都有衛星。Ali 與 Zhang 報告這使 motif 需要完整部署，與部分填滿的真實 Shell 1 各有條件，而他們在 90 天 TLE 上的穩定鏈路設計達到 52.9 ms 與 8.2 跳，調整後的 +Grid 則為 89.0 ms 與 23.4 跳preprint，所以 motif 結果最好被解讀為規則接線所能達到的上界。
- Setup delay motivates but is outside the model. The three-zone design behind the 54% headline changes satellite ISLs about every 12 minutes. The authors accept setup from a few seconds to small tens of seconds. Setup cost itself lies outside the metrics. [1, §5.4] Modern terminals require 2–30 s (Part C, item 18). Later work finds a noticeable latency increase from 100 ms setup per path change. Routing churn and the 15-second scheduler lie outside this model (Part C, items 56–57). [32, §IV-B]建鏈延遲是動機則位於成本模型之外。整個設計的存在就是為了避免重新對準，但 54% 這個主要數字背後的三區多 motif 會讓每顆衛星約每 12 分鐘更換一次 ISL，作者認為對數秒到十幾秒的建鏈時間而言可以容忍，其成本則位於指標範圍之外[1, §5.4]。現代終端需要 2–30 秒（C 部第 18 項），後續研究顯示每次路徑改變 100 ms 的建鏈成本會明顯提高平均延遲[32, §IV-B]，而路由層的變動或 15 秒排程器皆位於模型範圍之外（C 部第 56–57 項）。
- Capacity is a hop-count proxy. The "nearly 2×" throughput is inferred from fewer ISL hops per connection rather than computed as a flow. The paper uses link rate outside the model, only noting that 5.6 Gbps laser ISLs over 4,900 km had already been demonstrated [1, §2.3–2.4]. Applying it to the measured network requires a separate evaluation, whose ISL paths are dominated by routing policy rather than topology (Part C, item 58).容量是以跳數代理。「近 2 倍」的吞吐量是從每條連線經過較少 ISL 跳數推得，其依據為拓樸代理指標，論文將鏈路速率保留為待定，只提到 4,900 公里、5.6 Gbps 的雷射 ISL 已經實現過[1, §2.3–2.4]。它套用到實測網路時須另行評估，後者的 ISL 路徑由路由政策主導，拓樸則須配合路由條件解讀（C 部第 58 項）。
- Numbers to read with care. Percentages combine median M1 and traffic-weighted Φ1. They also mix a single motif, gaining about 44% on 40×40, with the Table 1 three-zone design. The 40×40 baseline is one snapshot. Minimum ranges come from +Grid connectivity, providing a worst case rather than terminal specifications. [1, §4.3–4.4, §6]需小心閱讀的數字。主要百分比混用了兩種指標，即城市對的中位 M1 與流量加權的 Φ1，也混用了兩種設計，即在 40×40 上約 44% 的單一 motif 與表 1 的三區多 motif，而 40×40 基準只是單一快照，最小鏈路範圍則是 +Grid 仍能連通的最短範圍，被當作最壞情況其依據為拓樸可行條件[1, §4.3–4.4, §6]。
7. How it connects7. 與其他內容的關聯
- Section 5.2 builds on the 54% result and places it beside Ali and Zhang's real-ephemeris topologies and TinyLEO's three-ISL baseline.第 5.2 節以 54% 的結果為基礎，並將它與 Ali 和 Zhang 的真實星曆拓樸以及 TinyLEO 的三 ISL 基準並列。
- Section 5.7 tabulates the 24×66 layout, four terminals and unmodeled setup delay among the changed assumptions.第 5.7 節把 24×66 佈局、四個終端與模型範圍之外的建鏈延遲列入已改變的假設。
- Section 5.1 explains the stretch metric that the motif objective minimizes, defined there as a term.第 5.1 節解釋 motif 目標所最小化的拉伸度量，並在該處定義為術語。
- Section 2.5 explains +Grid, the 5,014 km visibility limit and motif gains. Section 2.4 covers acquisition time and its churn cost.第 2.5 節在雷射鏈路的脈絡中給出 +Grid、5,014 公里可見性極限與 motif 的改善，第 2.4 節說明使變動代價高昂的對準時間。
- Sibling digests: Gearing Up 2018 is the agenda paper this one answers. Handley 2018 inferred the lattice the +Grid baseline formalises. CCR 2020 and Hypatia adopt the +Grid as their default. Temporary laser ISLs and Setup delay quantify the churn cost motifs avoid. Ali & Zhang 2026 and TinyLEO are the three-terminal, real-data successors.相關導讀：Gearing Up 2018 是這篇所回應的議程論文。Handley 2018 推斷出 +Grid 基準所形式化的格狀結構。CCR 2020 與 Hypatia 採用 +Grid 作為預設。暫時雷射 ISL 與 建鏈延遲量化 motif 所避免的變動成本。Ali 與 Zhang 2026 和 TinyLEO 是三終端、真實資料的後繼者。
Key numbers關鍵數字
| Number | 數字 | Meaning (setup) | 意義（設定） |
|---|---|---|---|
| 54% / 40% | Φ1 improvement of the three-zone multi-motif over +Grid for Starlink 24×66 at 5,014 km / 2,006 km link range [1, Tab. 1]. | Starlink 24×66 在 5,014 公里／2,006 公里鏈路範圍下，三區多 motif 相對 +Grid 的 Φ1 改善[1, Tab. 1]。 | |
| 45% / 4% | Same for Kuiper phase A (34×34, 1,156 satellites, 630 km, 51.9°) at 5,440 km / 1,761 km [1, §6, Tab. 1]. | Kuiper A 階段（34×34，1,156 顆，630 公里，51.9°）在 5,440 公里／1,761 公里下的同一度量[1, §6, Tab. 1]。 | |
| 48% / 7% | Same for the 40×40 reference constellation at 5,014 km / its minimum range [1, Tab. 1]. | 40×40 參考星座在 5,014 公里／其最小範圍下的同一度量[1, Tab. 1]。 | |
| 1.25 / 10.57 | +Grid population-weighted average stretch and hop count on the 40×40 constellation, one snapshot [1, §4.3]. | 40×40 星座上 +Grid 的人口加權平均拉伸與跳數，單一快照[1, §4.3]。 | |
| 24 × 66, 550 km, 53° | Starlink Phase 1 as filed in November 2018, 1,584 satellites, four ISLs each [1, §3.1, §6]. Superseded by 72 × 22 and three terminals. | 2018 年 11 月申請的 Starlink 第一階段，1,584 顆、各四條 ISL[1, §3.1, §6]。已被 72×22 與三個終端取代。 | |
| 5,014 / 2,006 km | Visibility-limited maximum and +Grid-minimum (worst-case power-limited) ISL range for Starlink [1, §2.3, §6]. 16.7 ms and 6.69 ms one way [computed]. | Starlink 的可見性限制最大 ISL 範圍與 +Grid 最小範圍（功率受限最壞情況）[1, §2.3, §6]。單向 16.7 ms 與 6.69 ms［計算值］。 | |
| 9% / 19% | Links shared by two ILP topologies computed one minute apart / random-graph links infeasible within five minutes [1, §3.1–3.2]. | 相隔一分鐘算出的兩個 ILP 拓樸共用的鏈路比例／隨機圖中五分鐘內變得超出可行條件的鏈路比例[1, §3.1–3.2]。 | |
| ≈ 1,100 → 3,600 | A 53°-inclined 40×40 constellation has the listed Equator and 53° candidate counts. A polar constellation rises from about 600 to 4,000 candidates at 90°. [1, §5.1, Fig. 12] | 53° 傾角 40×40 星座在赤道與 53° 處可能的 motif 數。極軌道星座則從約 600 個增加到 90° 的約 4,000 個[1, §5.1, Fig. 12]。 | |
| 48.9 vs 60.6 ms | Balanced motif versus +Grid on a synthetic full 72×22 shell with four ISLs [34, Tab. 2] [preprint]. | 合成完整 72×22 殼層、四條 ISL 下平衡 motif 對比 +Grid[34, Tab. 2]［預印本］。 |