pith. sign in

arxiv: 2510.25498 · v2 · submitted 2025-10-29 · 💻 cs.NI · cs.PF

Evaluating Learning Congestion control Schemes for LEO Constellations

Mihai Mazilu , Aiden Valentine , George Parisis This is my paper

Pith reviewed 2026-05-18 03:31 UTC · model grok-4.3

classification 💻 cs.NI cs.PF
keywords LEO satellite networkscongestion controlreinforcement learninghandoversRTT dynamicsnetwork emulationactive queue management
0
0 comments X

The pith

Reinforcement learning congestion control schemes severely underperform under LEO dynamic conditions despite resisting non-congestive loss.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper evaluates representative congestion control algorithms from loss-based, model-based, and learning-based classes in LEO satellite networks using the LeoEM framework for orbital dynamics combined with Mininet micro-benchmarks. It shows that handover-aware loss-based schemes reclaim bandwidth at higher latency cost, BBRv3 holds high throughput with slow reactions to RTT shifts, and RL schemes like Vivace, Sage and Astraea lag in adapting to handovers and changing round-trip times. Fairness suffers under RTT asymmetry and multiple bottlenecks but improves with active queue management. A sympathetic reader cares because these limitations point to the need for transport protocols tailored to LEO characteristics rather than relying on terrestrial designs.

Core claim

The paper claims that RL-based congestion control schemes severely underperform under dynamic LEO conditions of frequent handovers and rapidly changing RTTs, even though they prove notably resistant to non-congestive loss, while exposing critical limitations across current schemes and offering design insights for LEO-specific data transport protocols.

What carries the argument

LeoEM framework combined with Mininet micro-benchmarks that emulate realistic LEO orbital dynamics, handover behavior, and non-congestive loss patterns for testing single-flow, multi-flow, and AQM scenarios.

If this is right

  • Handover-aware loss-based schemes can reclaim bandwidth but increase latency.
  • BBRv3 sustains high throughput with modest delay penalties yet reacts slowly to abrupt RTT changes.
  • RL-based schemes severely underperform under dynamic conditions despite resistance to non-congestive loss.
  • Fairness degrades with RTT asymmetry and multiple bottlenecks especially in human-designed schemes.
  • AQM at bottlenecks can restore fairness and boost efficiency.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • Future LEO protocol designs could benefit from hybrid mechanisms that add explicit handover prediction to loss-based or model-based schemes.
  • The resistance of RL methods to non-congestive loss suggests they might become viable with retraining on LEO-specific traces that include predictable handover sequences.
  • This evaluation implies that multi-bottleneck fairness tests should become standard for any new CC algorithm targeting satellite networks.

Load-bearing premise

The LeoEM emulation with Mininet sufficiently captures real-world LEO orbital dynamics, handover behavior, and non-congestive loss so that observed performance differences generalize beyond the tested scenarios.

What would settle it

A field deployment or trace-driven test on live LEO satellites where RL-based schemes achieve throughput and fairness comparable to or better than BBRv3 and loss-based schemes under rapid handovers would falsify the underperformance claim.

Figures

Figures reproduced from arXiv: 2510.25498 by Aiden Valentine, George Parisis, Mihai Mazilu.

Figure 1
Figure 1. Figure 1: LeoEM Experimentation with a single flow (a) and (b), and two competing flows (c) [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Goodput evolution of two competing flows (solid/dashed lines) and base RTT of the SD to NY path (red dotted line). [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Responsiveness. Cumulative Distribution of Goodput A. Responsiveness In this section, we study how the selected CC approaches react to dynamically changing network conditions similar to the ones encountered in Section III (e.g. due to link re-configuration and routing changes, or transient hotspots). Through our micro-benchmarks, we can explore how CC approaches behave in the presence of changes in bandwid… view at source ↗
Figure 4
Figure 4. Figure 4: Responsiveness. Sending Rate in Time 50 100 150 200 RTT (ms) 0.0 0.5 1.0 Goodput Ratio Cubic BBRv3 Vivace Sage Astraea [PITH_FULL_IMAGE:figures/full_fig_p006_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Goodput ratio of two competing flows on an emulated [PITH_FULL_IMAGE:figures/full_fig_p006_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Goodput ratio of two competing flows flows on an emulated LEO satellite path. Starting flow has 20 ms RTT and the [PITH_FULL_IMAGE:figures/full_fig_p007_6.png] view at source ↗
Figure 8
Figure 8. Figure 8: Path Reconfiguration: Normalised aggregate goodput vs Jain’s fairness index. paths are 20ms and 50ms, respectively. Initially, we start 2 flows in each of those paths. Path reconfiguration is emulated by redirecting traffic from the second to the first path (100 seconds into the experiment) and then back to the original path (200 seconds into the experiment). These changes are made in place so no packets a… view at source ↗
Figure 9
Figure 9. Figure 9: AQM Performance. (a) Inter-RTT experiment with SFQ at the bottleneck, (b) three-bottleneck experiment with SFQ, and (c) efficiency of 5 competing flows under different AQM schemes, comparing normalised throughput and delay inflation. In the multi-bottleneck scenario (Figure 9b), it is clear that fair queueing enforces max-min fairness. We observe a similar unfairness pattern with Sage as the base RTT incre… view at source ↗
read the original abstract

Low Earth Orbit (LEO) satellite networks introduce unique congestion control (CC) challenges due to frequent handovers, rapidly changing round-trip times (RTTs), and non-congestive loss. This paper presents the first comprehensive, emulation-driven evaluation of CC schemes in LEO networks, combining realistic orbital dynamics via the LeoEM framework with targeted Mininet micro-benchmarks. We evaluated representative CC algorithms from three classes, loss-based (Cubic, SaTCP), model-based (BBRv3), and learning-based (Vivace, Sage, Astraea), across diverse single-flow and multi-flow scenarios, including interactions with active queue management (AQM). Our findings reveal that: (1) handover-aware loss-based schemes can reclaim bandwidth but at the cost of increased latency; (2) BBRv3 sustains high throughput with modest delay penalties, yet reacts slowly to abrupt RTT changes; (3) RL-based schemes severely underperform under dynamic conditions, despite being notably resistant to non-congestive loss; (4) fairness degrades significantly with RTT asymmetry and multiple bottlenecks, especially in human-designed CC schemes; and (5) AQM at bottlenecks can restore fairness and boost efficiency. These results expose critical limitations in current CC schemes and provide insight for designing LEO-specific data transport protocols.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. The manuscript presents an emulation-based evaluation of congestion control schemes for LEO satellite networks. It combines the LeoEM framework for modeling orbital dynamics and handovers with Mininet micro-benchmarks to compare loss-based (Cubic, SaTCP), model-based (BBRv3), and learning-based (Vivace, Sage, Astraea) algorithms across single-flow and multi-flow scenarios, including RTT asymmetry, multiple bottlenecks, non-congestive loss, and AQM interactions. The central claims are that RL-based schemes severely underperform under dynamic LEO conditions despite resistance to non-congestive loss, BBRv3 sustains throughput with modest delay, handover-aware loss-based schemes trade latency for bandwidth, fairness degrades with asymmetry, and AQM can restore efficiency.

Significance. If the emulation results generalize, the work provides valuable empirical insights into limitations of current CC schemes in high-mobility LEO environments and useful guidance for LEO-specific protocol design. The multi-class comparison and inclusion of AQM and multi-flow interactions are strengths. The approach credits the use of a realistic orbital model (LeoEM) rather than simplified assumptions.

major comments (2)
  1. [§3] §3 (Emulation Setup): The LeoEM+Mininet framework is presented as capturing relevant LEO dynamics, but the manuscript provides no quantitative validation (e.g., Kolmogorov-Smirnov distance or moment matching) between simulated RTT/handover traces and measurements from operational LEO systems. This is load-bearing for the claim that RL schemes inherently underperform, as smoother simulated dynamics could artifactually produce the reported throughput collapse and slow reaction.
  2. [§5] §5 (Results): Performance plots for RL schemes (e.g., throughput and delay under dynamic RTT) report point estimates without error bars, number of runs, or statistical tests. This makes it difficult to determine whether the 'severe underperformance' relative to BBRv3 is robust or sensitive to scenario parameterization.
minor comments (2)
  1. [Abstract] Abstract and §2: The claim of being the 'first comprehensive' evaluation should be tempered with citations to prior LEO CC studies to avoid overstatement.
  2. [§5] Figure captions in §5: Several figures would benefit from explicit legends listing all CC schemes and scenario parameters for readability.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive and detailed comments on our manuscript. We address each major comment below and indicate the revisions we will make to strengthen the paper.

read point-by-point responses

  1. Referee: [§3] §3 (Emulation Setup): The LeoEM+Mininet framework is presented as capturing relevant LEO dynamics, but the manuscript provides no quantitative validation (e.g., Kolmogorov-Smirnov distance or moment matching) between simulated RTT/handover traces and measurements from operational LEO systems. This is load-bearing for the claim that RL schemes inherently underperform, as smoother simulated dynamics could artifactually produce the reported throughput collapse and slow reaction.

    Authors: We acknowledge the value of quantitative validation for the emulation framework. LeoEM relies on publicly documented orbital mechanics and ephemeris data with standard propagation models; however, direct access to proprietary RTT and handover traces from operational LEO constellations is not available for statistical tests such as Kolmogorov-Smirnov distance. In the revision we will expand §3 with an explicit discussion of the model's sources, references to existing validations of comparable LEO emulators in the literature, and a clear statement of remaining limitations. This will better contextualize the applicability of our results without overstating the framework's fidelity. revision: partial

  2. Referee: [§5] §5 (Results): Performance plots for RL schemes (e.g., throughput and delay under dynamic RTT) report point estimates without error bars, number of runs, or statistical tests. This makes it difficult to determine whether the 'severe underperformance' relative to BBRv3 is robust or sensitive to scenario parameterization.

    Authors: We agree that the absence of statistical information limits interpretability. We will rerun the key experiments with a minimum of ten independent trials per scenario, add error bars (standard deviation) to all relevant plots in §5, and include a methods paragraph specifying the number of runs together with any statistical comparisons performed between schemes. revision: yes

Circularity Check

0 steps flagged

No circularity: empirical evaluation with external frameworks

full rationale

This paper is a simulation-based empirical study that evaluates existing congestion control algorithms (Cubic, BBRv3, Vivace, etc.) inside the LeoEM orbital model combined with Mininet. No equations, first-principles derivations, or predictions are presented that could reduce to fitted parameters or self-referential definitions. All reported performance differences arise from direct measurement of throughput, latency, and fairness under controlled emulation scenarios; the central claims therefore rest on observable experimental outcomes rather than any construction that equates outputs to inputs by definition.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

Evaluation depends on the fidelity of the LeoEM orbital model and the assumption that selected scenarios represent typical LEO traffic patterns; no free parameters or invented entities are introduced in the abstract.

axioms (2)
  • domain assumption LeoEM framework accurately reproduces LEO orbital dynamics, handover events, and RTT variations.
    Central to the emulation setup described in the abstract.
  • domain assumption The chosen representative algorithms (Cubic, SaTCP, BBRv3, Vivace, Sage, Astraea) adequately cover the space of loss-based, model-based, and learning-based CC schemes.
    Used to generalize findings to broader classes of algorithms.

pith-pipeline@v0.9.0 · 5758 in / 1400 out tokens · 37860 ms · 2026-05-18T03:31:33.218213+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

42 extracted references · 42 canonical work pages · 1 internal anchor

  1. [1]

    Gearing up for the 21st century space race,

    D. Bhattacherjeeet al., “Gearing up for the 21st century space race,” in Proc. of HotNets, 2018

    work page 2018

  2. [2]

    Delay is not an option: Low latency routing in space,

    M. Handley, “Delay is not an option: Low latency routing in space,” in Proc. of HotNets, 2018

    work page 2018

  3. [3]

    CUBIC: a new TCP-friendly high-speed TCP variant,

    S. Ha, I. Rhee, and L. Xu, “CUBIC: a new TCP-friendly high-speed TCP variant,”SIGOPS Oper. Syst. Rev., p. 64–74, 2008

    work page 2008

  4. [4]

    Exploring the “Internet from Space

    S. Kassing, D. Bhattacherjee, A. B. ´Aguas, J. E. Saethre, and A. Singla, “Exploring the “Internet from Space” with Hypatia,” inProc. of ACM IMC, 2020

    work page 2020

  5. [5]

    Reinforcement learning-based congestion control: A systematic evaluation of fairness, efficiency and responsive- ness,

    L. Giacomoni and G. Parisis, “Reinforcement learning-based congestion control: A systematic evaluation of fairness, efficiency and responsive- ness,” inProc of IEEE INFOCOM, 2024

    work page 2024

  6. [6]

    Experimental evaluation of BBR congestion control,

    M. Hock, R. Bless, and M. Zitterbart, “Experimental evaluation of BBR congestion control,” inProc of ICNP, 2017

    work page 2017

  7. [7]

    Fade Slope Analysis of Ka-Band Earth-LEO Satellite Links Using a Synthetic Rain Field Model,

    W. Liu and D. G. Michelson, “Fade Slope Analysis of Ka-Band Earth-LEO Satellite Links Using a Synthetic Rain Field Model,”IEEE Transactions on Vehicular Technology, vol. 58, no. 8, 2009

    work page 2009

  8. [8]

    Network topology design at 27,000 km/hour,

    D. Bhattacherjee and A. Singla, “Network topology design at 27,000 km/hour,” inProc. of CoNEXT, 2019

    work page 2019

  9. [9]

    SaTCP: Link-Layer Informed TCP Adaptation for Highly Dynamic LEO Satellite Networks,

    X. Cao and X. Zhang, “SaTCP: Link-Layer Informed TCP Adaptation for Highly Dynamic LEO Satellite Networks,” inProc. of IEEE INFO- COM, 2023

    work page 2023

  10. [10]

    Developing and experimenting with LEO satellite constellations in OMNeT++,

    A. Valentine and G. Parisis, “Developing and experimenting with LEO satellite constellations in OMNeT++,” 2021. [Online]. Available: https://arxiv.org/abs/2109.12046

    work page arXiv 2021

  11. [11]

    Network Characteristics of LEO Satellite Constellations: A Starlink-Based Mea- surement from End Users,

    S. Ma, Y . C. Chou, H. Zhao, L. Chen, X. Ma, and J. Liu, “Network Characteristics of LEO Satellite Constellations: A Starlink-Based Mea- surement from End Users,” inProc of IEEE INFOCOM, 2023

    work page 2023

  12. [12]

    A Distributed Congestion Control Routing Protocol Based on Traffic Classification in LEO Satellite Networks,

    S. Dai, L. Rui, S. Chen, and X. Qiu, “A Distributed Congestion Control Routing Protocol Based on Traffic Classification in LEO Satellite Networks,” inProc. of IFIP/IEEE IM, 2021

    work page 2021

  13. [13]

    Experimental evaluation of TCP protocols for high-speed networks,

    Y .-T. Li, D. Leith, and R. N. Shorten, “Experimental evaluation of TCP protocols for high-speed networks,”IEEE/ACM Transactions on Networking, vol. 15, no. 5, 2007

    work page 2007

  14. [14]

    A Compar- ative Evaluation of TCP Congestion Control Schemes over Low-Earth- Orbit (LEO) Satellite Networks,

    G. Barbosa, S. Theeranantachai, B. Zhang, and L. Zhang, “A Compar- ative Evaluation of TCP Congestion Control Schemes over Low-Earth- Orbit (LEO) Satellite Networks,” inProc of AINTEC, 2023

    work page 2023

  15. [15]

    StarQUIC: Tuning Congestion Control Algorithms for QUIC over LEO Satellite Networks,

    V . Kamel, J. Zhao, D. Li, and J. Pan, “StarQUIC: Tuning Congestion Control Algorithms for QUIC over LEO Satellite Networks,” inProc of LEO-NET, 2024

    work page 2024

  16. [16]

    BBRv3: Algorithm Bug Fixes and Public Internet Deployment,

    N. Cardwellet al., “BBRv3: Algorithm Bug Fixes and Public Internet Deployment,” 2023. [Online]. Avail- able: https://datatracker.ietf.org/meeting/117/materials/slides-117-ccwg- bbrv3-algorithm-bug-fixes-and-public-internet-deployment-00

    work page 2023

  17. [17]

    PCC Vivace: Online-Learning Congestion Control,

    M. Donget al., “PCC Vivace: Online-Learning Congestion Control,” in Proc. of USENIX NSDI, 2018

    work page 2018

  18. [18]

    Computers Can Learn from the Heuristic Designs and Master Internet Congestion Control,

    C.-Y . Yen, S. Abbasloo, and H. J. Chao, “Computers Can Learn from the Heuristic Designs and Master Internet Congestion Control,” inProc. of ACM SIGCOMM, 2023

    work page 2023

  19. [19]

    Astraea: Towards fair and efficient learning-based congestion control,

    X. Liao, H. Tian, C. Zeng, X. Wan, and K. Chen, “Astraea: Towards fair and efficient learning-based congestion control,” inProc. of EuroSys, 2024

    work page 2024

  20. [20]

    Mininet

    “Mininet.” [Online]. Available: http://mininet.org/

    work page

  21. [21]

    Performance and improvements of tcp cubic in low-delay cellular networks,

    P. Bruhn, M. K ¨uhlewind, and M. Muehleisen, “Performance and improvements of tcp cubic in low-delay cellular networks,” Computer Networks, vol. 224, p. 109609, 2023. [Online]. Available: https://www.sciencedirect.com/science/article/pii/S1389128623000543

    work page 2023

  22. [22]

    BBR: Congestion-based congestion control,

    N. Cardwell, Y . Cheng, C. S. Gunn, S. H. Yeganeh, and V . Jacobson, “BBR: Congestion-based congestion control,”ACM Queue, pp. 20 – 53, 2016

    work page 2016

  23. [23]

    Classic Meets Modern: a Pragmatic Learning-Based Congestion Control for the Internet,

    S. Abbaslooet al., “Classic Meets Modern: a Pragmatic Learning-Based Congestion Control for the Internet,” inProc. of ACM SIGCOMM, 2020

    work page 2020

  24. [24]

    BBR3 enabled kernel

    “BBR3 enabled kernel.” [Online]. Available: https://github.com/google/bbr/tree/v3

    work page

  25. [25]

    Hyper-v technology overview

    “Hyper-v technology overview.” [Online]. Available: https://learn.microsoft.com/en-us/windows-server/virtualization/hyper- v/hyper-v-overview?pivots=windows

    work page

  26. [26]

    Code and dataset for the paper titled taming congestion in space: A study of congestion control on leo networks

    “Code and dataset for the paper titled taming congestion in space: A study of congestion control on leo networks.” [Online]. Available: https://figshare.com/s/32821d6f5940f72fd946

    work page

  27. [27]

    Available: https://iperf.fr/

    “iperf.” [Online]. Available: https://iperf.fr/

    work page

  28. [28]

    Udt documentation

    “Udt documentation.” [Online]. Available: https://udt.sourceforge.io/udt3/

    work page

  29. [29]

    ss(8) — linux manual page

    “ss(8) — linux manual page.” [Online]. Available: https://man7.org/linux/man-pages/man8/ss.8.html

    work page

  30. [30]

    Towards a Deeper Understanding of TCP BBR Congestion Control,

    D. Scholzet al., “Towards a Deeper Understanding of TCP BBR Congestion Control,” inProc. of IFIP Networking, 2018

    work page 2018

  31. [31]

    A Multifaceted Look at Starlink Performance,

    N. Mohanet al., “A Multifaceted Look at Starlink Performance,” in Proc of the ACM Web Conference, 2024

    work page 2024

  32. [32]

    Differentiated end-to-end Internet services using a weighted proportional fair sharing TCP,

    J. Crowcroft and P. Oechslin, “Differentiated end-to-end Internet services using a weighted proportional fair sharing TCP,”SIGCOMM Comput. Commun. Rev., 1998

    work page 1998

  33. [33]

    Piece of CAKE: A Comprehensive Queue Management Solution for Home Gateways

    T. Høiland-Jørgensen, D. T ¨aht, and J. Morton, “Piece of cake: A comprehensive queue management solution for home gateways,” 2018. [Online]. Available: https://arxiv.org/abs/1804.07617

    work page internal anchor Pith review Pith/arXiv arXiv 2018

  34. [34]

    Improving Starlink’s Latency,

    “Improving Starlink’s Latency,” Online, 2022. [Online]. Available: https://www.starlink.com/public-files/StarlinkLatency.pdf

    work page 2022

  35. [35]

    Scouting the path to a million-client server,

    Y . Zhao, A. Saeed, M. Ammar, and E. Zegura, “Scouting the path to a million-client server,” inPassive and Active Measurement. Springer International Publishing, 2021, pp. 337–354

    work page 2021

  36. [36]

    Bounding queue delay in cellular networks to support ultra-low latency applications,

    S. Abbasloo and H. J. Chao, “Bounding queue delay in cellular networks to support ultra-low latency applications,” 2019. [Online]. Available: https://arxiv.org/abs/1908.00953

    work page arXiv 2019

  37. [37]

    The Flow Queue CoDel Packet Scheduler and Active Queue Management Algorithm,

    T. Høiland-Jørgensen, P. McKenney, dave.taht@gmail.com, J. Gettys, and E. Dumazet, “The Flow Queue CoDel Packet Scheduler and Active Queue Management Algorithm,” RFC 8290, Jan. 2018. [Online]. Available: https://www.rfc-editor.org/info/rfc8290

    work page 2018

  38. [38]

    Fq-pie queue discipline in the linux kernel: Design, implementation and challenges,

    G. Ramakrishnan, M. Bhasi, V . Saicharan, L. Monis, S. D. Patil, and M. P. Tahiliani, “Fq-pie queue discipline in the linux kernel: Design, implementation and challenges,” in2019 IEEE 44th LCN Symposium on Emerging Topics in Networking (LCN Symposium), 2019, pp. 117– 124

    work page 2019

  39. [39]

    Design and evaluation of cobalt queue discipline,

    J. Palmei, S. Gupta, P. Imputato, J. Morton, M. P. Tahiliani, S. Avallone, and D. T ¨aht, “Design and evaluation of cobalt queue discipline,” in 2019 IEEE International Symposium on Local and Metropolitan Area Networks (LANMAN), 2019, pp. 1–6

    work page 2019

  40. [40]

    StarryNet: Empowering researchers to evaluate futuristic integrated space and terrestrial networks,

    Z. Lai, H. Liet al., “StarryNet: Empowering researchers to evaluate futuristic integrated space and terrestrial networks,” in20th USENIX Symposium on Networked Systems Design and Implementation (NSDI 23). Boston, MA: USENIX Association, Apr. 2023, pp. 1309–1324. [Online]. Available: https://www.usenix.org/conference/nsdi23/presentation/lai-zeqi

    work page 2023

  41. [41]

    xeoverse: A real-time simulation platform for large leo satellite mega-constellations,

    M. M. Kassem and N. Sastry, “xeoverse: A real-time simulation platform for large leo satellite mega-constellations,” 2024. [Online]. Available: https://arxiv.org/abs/2406.11366

    work page arXiv 2024

  42. [42]

    Tcp congestion control performance over starlink,

    J. Garcia, S. Sundberg, and A. Brunstrom, “Tcp congestion control performance over starlink,” inProceedings of the 2025 Applied Networking Research Workshop, ser. ANRW ’25. New York, NY , USA: Association for Computing Machinery, 2025. [Online]. Available: https://doi.org/10.1145/3744200.3744760

    work page doi:10.1145/3744200.3744760 2025