PaperlayerView exact revision on arXiv

Source-linked AI summary

Network Characteristics of LEO Satellite Constellations: A Starlink-Based Measurement from End Users

Sami Ma, Yi Ching Chou, Haoyuan Zhao, Long Chen, Xiaoqiang Ma, Jiangchuan Liu

arXiv:2212.13697v1cs.NI

TL;DR

Large-scale LSN performance lacks comprehensive real-world evidence, so the paper systematically measures Starlink from end users across diverse scenarios and applications. The study finds more dynamic throughput and latency, frequent outages, environmental sensitivity, bent-pipe limitations, and distinct mobility and remote-area challenges.

  • Problem

    Few comprehensive studies provide deep insight into the practical performance of large-scale LSNs, despite their proposed role in global Internet coverage and 6G.

  • Method

    The paper conducts a systematic end-user measurement study of Starlink using diverse configurations, applications, environments, and remote-area deployments.

  • Results

    Starlink users experience more dynamic throughput and latency than terrestrial users, frequent outages, environmental sensitivity, and bent-pipe-related network limitations.

  • Takeaways & Limitations

    The measurements support LSNs as promising for ubiquitous Internet coverage while identifying issues relevant to optimizing deployment, operation, networks, and applications.

Abstract

from arXiv · show

Low Earth orbit Satellite Networks (LSNs) have been advocated as a key infrastructure for truly global coverage in the forthcoming 6G. This paper presents our initial measurement results and observations on the end-to-end network characteristics of Starlink, arguably the largest LSN constellation to date. Our findings confirm that LSNs are a promising solution towards ubiquitous Internet coverage over the Earth; yet, we also find that the users of Starlink experience much more dynamics in throughput and latency than terrestrial network users, and even frequent outages. Its user experiences are heavily affected by environmental factors such as terrain, solar storms, rain, clouds, and temperature, so is the power consumption. We further analyze Starlink's current bent-pipe relay strategy and its limits, particularly for cross-ocean routes. We have also explored its mobility and portability potentials, and extended our experiments from urban cities to wild remote areas that are facing distinct practical and cultural challenges.

I. INTRODUCTION

The paper conducts a systematic, end-user measurement study of Starlink to assess performance, influential factors, coverage, mobility, and remote-area challenges. It finds dynamic service, frequent outages, environmental sensitivity, and limitations associated with Starlink’s current network strategy.

  • Few comprehensive real-world studies characterize large-scale LSN performance, making systematic measurement necessary despite complex, dynamic, and heterogeneous architectures.
  • Systematic Starlink measurements target end-to-end performance from the perspective of common users and diverse configurations and applications.The study focuses on Starlink as a large LSN constellation and plug-and-play black-box service.
  • The study asks whether Starlink can match modern terrestrial performance, which factors influence user experience, and whether its constellation provides global coverage.
  • Starlink throughput and latency are more dynamic than terrestrial networks, with frequent outages reported in the measurements.
  • Terrain, solar storms, rain, clouds, and temperature can heavily affect Starlink user experience and power consumption.
  • Mobility maintains throughput relatively well but still involves frequent outages and high latency, while remote-area tests expose practical and cultural challenges.

II. RELATED WORK

Prior work spans satellite theory, system design, deployment, simulations, and limited real-world Starlink studies. This paper positions itself as a broader end-user measurement effort covering global endpoints, video streaming, remote regions, and power consumption.

  • Satellite-network research includes theoretical analysis, system design, practical deployment, and studies of topology, routing, ground links, rain, and solar storms.
  • Real-world Starlink, OneWeb, and Telesat deployments have motivated many reviews, while LEO studies have largely relied on simulators such as Hypatia and StarPerf.
  • Simulators support model and design validation, but real-world measurement is described as irreplaceable for large-scale complex systems such as Starlink.
  • Earlier Starlink measurements examined selected performance aspects, whereas this work measures common end users with more global coverage and additional applications and settings.The added scope includes distant endpoints, video streaming, remote wild northern areas, and power consumption.
  • Starlink comprises LEO satellites, ground stations, and user terminals, with over 2,000 satellites and more than 400,000 subscribers described in the paper.Most deployed satellites are in the first shell, consisting of 72 orbits at 550 km with 53° inclination.
  • Consumer Starlink uses a dish to connect through a User Link to a visible satellite, with Gen 1 and Gen 2 consumer dishes differing in size, weight, and router bandwidth streams.

B. Measurement Topology and Environment

The measurement study deployed four dishes across urban and remote environments and evaluated multiple protocols, applications, tools, and worldwide endpoints. Terrestrial cable service provided the comparison baseline for applicable experiments.

  • Four dish kits were monitored for seven months beginning in January 2022, including three Gen 1 dishes and one more portable Gen 2 dish.
  • Experiments covered Vancouver-area cities, remote steep valleys, and far-northern locations, spanning substantially different terrains.
  • A terrestrial cable service with a maximum download speed of 800 Mb/s served as the baseline, alongside the same AWS servers for comparison.
  • Measurements included Web browsing, file transfer, TCP and UDP throughput with iperf3, SCP transfers, and high-demand video streaming.
  • Latency was measured across nine AWS regions worldwide, with figures presenting global cumulative latency distributions and regional averages with standard deviations.
  • Starlink Mobile App data and YouTube’s Stats for nerds were collected through Python scripts that automatically launched tests and gathered outputs.

IV. URBAN CITY MEASUREMENT: STARLINK VERSUS TERRESTRIAL NETWORKING

Urban measurements compare Starlink with terrestrial networking across global endpoints, showing slightly higher but substantially less stable latency and routing patterns consistent with a single bent-pipe transmission.

  • 1) Latency across Regions:: Starlink latency is 10% higher on average than terrestrial networking but much more unstable across nine AWS regions.Potential influences include obstruction, satellite movement, and ISP routing decisions.
  • 1) Latency across Regions:: Latency gaps typically range from 1.8 to 22.8 ms, while Sydney is an outlier with gaps 3.4x to 43.6x larger.Bahrain also shows increased latency, although terrestrial networking experiences a similar increase there.
  • 1) Latency across Regions:: Starlink’s similar latency patterns across regions are consistent with routing through similar satellites, ground stations, and paths.The paper uses hourly latency patterns and the current routing framework to examine this behavior.
  • 1) Latency across Regions:: Cross-ocean communication can require switching from Starlink’s initial bent-pipe hop to terrestrial submarine-fiber routing.For Sydney, packets currently route through a Seattle AWS server after the first bent-pipe hop.

2) Latency Variation:

Starlink provides substantial throughput but exhibits greater variation, bursty losses, weaker TCP bandwidth utilization, and increasingly amplified transfer-time penalties as workloads grow.

  • 2) Latency Variation:: 3.8 times: Starlink’s latency variation is around 3.8 times that of the terrestrial network.Satellite movement and handovers are identified as likely causes of the common latency rises and falls.
  • 2) Latency Variation:: 0.255 ms download and 2.715 ms upload: Starlink jitter exceeds terrestrial jitter by more than 5x in both directions.Terrestrial jitter is 0.041 ms for download and 0.546 ms for upload.
  • 2) Latency Variation:: ∼80 Mb/s: Starlink achieves good average throughput, but its throughput standard deviation reaches 50.71% versus 34.44% terrestrially.The variation is attributed to changing satellite paths, handovers, and potentially low-cost terrestrial routing after Starlink traffic is passed onward.
  • 2) Latency Variation:: 0.24% download and 1.24% upload: bursty Starlink packet losses persist even at 20% of presented throughput.The low offered rate is intended to mitigate congestion-related explanations for packet loss.
  • 2) Latency Variation:: 39.0% versus 46.8%: Starlink’s TCP bandwidth utilization is lower than the terrestrial network’s.The results suggest TCP congestion control is sensitive to dynamics in the satellite hop; Gen 2 improves TCP uploads to 1.76x Gen 1, unlike UDP uploads at 0.73x.
  • 2) Latency Variation:: 200 MB: SCP transfer time over Starlink is almost double the terrestrial network’s, with higher variance as data size grows.The paper attributes the increasing penalty to accumulated satellite-hop dynamics.

C. Routing Strategy

Traceroute and regional latency correlations indicate that Starlink currently uses one bent-pipe communication before entering terrestrial routing, limiting direct dish-to-dish communication.

  • C. Routing Strategy: Starlink uses one bent-pipe communication along the measured route before switching to the terrestrial network.Traceroutes show a single SpaceX Services ISP entry at the ground station nearest the dish.
  • C. Routing Strategy: Moderate Pearson correlations across regions indicate similar Starlink latency patterns despite different destinations.The paper interprets this as evidence that similar satellites, ground stations, and routes serve those destinations.
  • C. Routing Strategy: The current architecture does not yet support nearby dishes exchanging data effectively through a satellite without ground stations or terrestrial nodes.The tunnel-based solution needs improvements for pair-wise dish communications.

D. Environmental Influential Factors

Starlink performance varies substantially with environmental conditions: obstructions, storms, precipitation, temperature, and dish power behavior all affect end-user operation.

  • Obstruction:: Transient surface obstructions can cause latency spikes or network outages, even when the location has a good visibility map.The paper distinguishes clear sky visibility from a clean dish surface as requirements for best performance.
  • Solar and Geomagnetic Storms:: Throughput dropped from 100 Mb/s to 5 Mb/s during February 3–4, 2022, coinciding with solar and geomagnetic storm events.Uploads remained relatively stable, suggesting asymmetric effects between directions during storms.
  • Precipitation, Temperature, and Dish Power:: 27%: average throughput loss during precipitation, with heavy rain above 4 mm per hour capping throughput.UDP downloads showed the strongest inverse correlation, while downloads fell from 215 Mb/s without precipitation to around 120 Mb/s at 4.1–5.2 mm of precipitation.
  • Precipitation, Temperature, and Dish Power:: 5–26%: throughput reduction when temperatures exceeded 12°C, while the dish relied on passive cooling through its aluminum back-plate.The lack of active cooling creates a challenge for summer or tropical operation.
  • Precipitation, Temperature, and Dish Power:: 56.3 Watts: average current-dish power consumption, with peaks reaching 144.5 Watts.Power consumption also rises persistently during rain or heavy clouds, likely to address interference.
  • Precipitation, Temperature, and Dish Power:: Power consumption cannot be measured per module because the dish kit is a black box, and it shows no obvious correlation with throughput.The authors observed inverse correlations in some upload and download cases.

V. STARLINK IN THE WILD

The study extends Starlink measurements beyond cities into remote areas without terrestrial Internet coverage, examining environments with distinct operational constraints.

  • STARLINK IN THE WILD: Remote-area experiments targeted locations lacking terrestrial Internet coverage, including cellular and power services.The study focuses on an estuary in the far north and a deep valley surrounded by steep mountains.

A. Northern Shoreline

In remote northern and mountainous settings, Starlink provided connectivity but with reduced performance, frequent interruptions, and substantial dependence on terrain, coverage, and available power.

  • Northern Shoreline: 68% lower: download throughput at Koeye Point than in the urban city, with latency 11% to 30% longer and intermittent interruptions.The authors associate this with fewer satellites and ground stations serving northern areas, plus handover and link-establishment demands.
  • High-Elevation Deep Valley: 13 Mb/s and 4 Mb/s: average download and upload throughput in Manning Park, with rare bursts to 100 Mb/s and 20 Mb/s.Latency averaged 90–350 ms, fluctuated above 1,000 ms, and outages occurred every 1–3 minutes.
  • High-Elevation Deep Valley: A reported 2% obstruction ratio in Manning Park did not prevent frequent outages because surrounding mountains could obstruct satellites outside the valley’s direct sky view.The comparison with an urban 24.9% obstruction ratio suggests that terrain and obstruction geometry matter more than the ratio alone.
  • Northern Shoreline: A 1,500 Ah power pack lasts around only 1 hour with the current Starlink kit in the off-grid remote locations.The estuary used a solar-diesel hybrid system, while the valley relied on a battery power pack.

VI. FURTHER DISCUSSIONS

The study examines bent-pipe stability through prolonged synchronous streaming, finding shared outages and user-dependent service differences even within one coverage area.

  • Stability of Bent-Pipe: Over 24 hours: two nearby dishes synchronously streamed an 8K YouTube video while their browser buffers were repeatedly cleaned for continuous downloading.The setup gave both dishes a shared ground-satellite link but separate user links.
  • Stability of Bent-Pipe: 37.5% for Dish A and 15.79% for Dish B: proportions of outage events attributed to their bent-pipe links.Overlapping outages between the dishes indicated a common bent-pipe-related factor.
  • Stability of Bent-Pipe: Different obstruction ratios of 2.7% and 4.7% coincided with different streaming stability between the two dishes in the same service area.Dish A was more stable, likely because it had clearer visibility than Dish B.

B. Mobility Potentials

Starlink supports connectivity while moving, but its mobile performance remains substantially less stable than stationary operation. Frequent outages, latency spikes, and boot-time delays limit practical mobility, while remote deployment introduces additional site constraints.

  • Mobility performance: The mobile dish may adjust its flat position during movement, supporting omnidirectional reception, but satellite switching may still be stressed in a valley.A six-minute straight drive in a valley coincided with a suspected need to switch satellites faster than expected.
  • Mobility performance: Frequent outages and high latency mean Starlink mobility still requires significantly more work despite throughput changing little during movement.The paper reports median outages every 16.5 seconds lasting 5 seconds, with some lasting up to 36 seconds.
  • Mobility performance: 100 ms average latency during motion reached 2,800 ms, with spikes above 200 ms occurring at least twice per minute on average.The authors associate some increases with dish orientation changes during turns or intermittent tree obstructions.
  • Mobility performance: 80 to 100 Mb/s download throughput during motion approached stationary averages but dropped sharply near the test’s end.Upload throughput also showed a noticeable drop, likely associated with a satellite disconnection.
  • Operational constraints: Starlink kits typically need 3-7 minutes to become Internet-ready, with observed cases exceeding 20 minutes, creating a barrier for mobile use.The in-motion test was conducted for academic research, and general use of Starlink dishes in motion was not then recommended.

C. Global Coverage? Technical and Cultural Implications

Starlink does not yet provide fully robust global coverage, and extending service into remote regions requires more satellites and inter-satellite links. Field deployment also encounters substantial environmental, power, wildlife, and cultural constraints.

  • Technical implications: Truly global coverage requires additional Starlink satellites, including polar shells, and inter-satellite links for efficient cross-continent communication and disaster resilience.The remote experiments were conducted 14° south of the Arctic Circle, nearly outside the current North American service area.
  • Practical implications: Remote rainforest deployment is constrained by 30-meter trees, limited clear sky, flooding, wildlife, and the absence of practical diesel power.The only clear sky was at a riverside weir subject to flooding, while bears, wolves, and birds could disrupt the dish.
  • Practical implications: A maximum dish demand of 145 watts makes intermittent operation inevitable when solar power is the only available source.This figure excludes snow-melting mode.
  • Cultural implications: Remote connectivity raises cultural concerns about pristine wilderness, electromagnetic signal pollution, distraction, and conflicts with Heiltsuk traditional values.The authors call for policies, professional guidelines, and ethics across disciplines and cultures.
  • Implications and future work: The measurements are intended to inform optimization of LSN deployment and operation and help developers and users customize networks and applications.Future work includes weather-attenuation models, mobility support, satellite-to-satellite links, and integration with terrestrial and 5G networks.
Loading 2212.13697v1…