Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Evaluating Learning Congestion control Schemes for LEO Constellations

Mihai Mazilu, Aiden Valentine, George Parisis

TL;DR

This work tackles the unique congestion-control challenges of LEO satellite networks, characterized by frequent handovers, rapid RTT variation, and non-congestive losses. It introduces an emulation-driven evaluation using LeoEM for orbital dynamics and Mininet micro-benchmarks to systematically compare loss-based Cubic/SaTCP, model-based BBRv3, and learning-based Vivace, Sage, and Astraea across single- and multi-flow scenarios and with AQMs. Key findings reveal that handover-aware loss-based schemes reclaim bandwidth at the cost of latency, BBRv3 maintains high throughput with moderate delay yet slow RTT adaptation, RL-based schemes under dynamic conditions underperform in bandwidth capture though Astraea improves fairness; AQM at bottlenecks can restore fairness and boost efficiency. These results identify critical limitations in current CC strategies and offer design directions for LEO-specific data transport protocols, highlighting the value of hybrid approaches and QoS-aware queue management for practical deployment.

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.

Evaluating Learning Congestion control Schemes for LEO Constellations

TL;DR

This work tackles the unique congestion-control challenges of LEO satellite networks, characterized by frequent handovers, rapid RTT variation, and non-congestive losses. It introduces an emulation-driven evaluation using LeoEM for orbital dynamics and Mininet micro-benchmarks to systematically compare loss-based Cubic/SaTCP, model-based BBRv3, and learning-based Vivace, Sage, and Astraea across single- and multi-flow scenarios and with AQMs. Key findings reveal that handover-aware loss-based schemes reclaim bandwidth at the cost of latency, BBRv3 maintains high throughput with moderate delay yet slow RTT adaptation, RL-based schemes under dynamic conditions underperform in bandwidth capture though Astraea improves fairness; AQM at bottlenecks can restore fairness and boost efficiency. These results identify critical limitations in current CC strategies and offer design directions for LEO-specific data transport protocols, highlighting the value of hybrid approaches and QoS-aware queue management for practical deployment.

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.

Paper Structure

This paper contains 20 sections, 9 figures.

Table of Contents

  1. Introduction
  2. Methodology
  3. Selected CC approaches
  4. Experimental setup
  5. Performance Metrics and Data Collection
  6. Experimentation with LeoEM
  7. Goodput
  8. Fairness
  9. Micro-benchmarks
  10. Responsiveness
  11. Fairness
  12. Intra-RTT fairness
  13. Inter-RTT fairness
  14. Multi-bottleneck fairness
  15. Path Reconfigurations
  16. ...and 5 more sections

Figures (9)

  • Figure 1: LeoEM Experimentation with a single flow (a) and (b), and two competing flows (c)
  • Figure 2: Goodput evolution of two competing flows (solid/dashed lines) and base RTT of the SD to NY path (red dotted line).
  • Figure 3: Responsiveness. Cumulative Distribution of Goodput
  • Figure 4: Responsiveness. Sending Rate in Time
  • Figure 5: Goodput ratio of two competing flows on an emulated LEO satellite path, both experiencing the same RTT.
  • ...and 4 more figures