TCP Congestion Control in Satellite Networks: An Evaluation of Algorithms and Policies

Evaluation of TCP Congestion Control Algorithms

Introduction

The study examines the performance of various TCP variants, specifically BBR, YeAH, and CUBIC, in the context of TCP congestion control for satellite communications. Emphasis is placed on scenarios utilizing Performance Enhancing Proxies (PEPs) and Active Queue Management (AQM) techniques, aimed at optimizing throughput while minimizing latency. Historically, satellite networks have suffered from degradation in TCP performance due to high latency and bandwidth limitations, prompting the need for optimized solutions.

Satellite Communication Landscape

  • Geostationary Orbit (GEO) Satellites: Typical applications of satellite communications occur in environments inaccessible to wired infrastructure, such as remote regions and mobile platforms (e.g., airplanes, ships). The shift towards Small GEO (SGEO) satellites, characterized by low cost and rapid deployment, signifies a change in the satellite industry’s focus.

  • Network Characteristics: Key challenges for satellite communications include high round-trip times (RTT), diverse packet corruption rates, and potential performance bottlenecks when using standard TCP variants.

TCP Variants Analysis

  • Introduction of TCP Variants: Variants such as Google’s BBR and satellite-optimized flavors have emerged to tackle inefficient TCP behavior over long-latency links. However, comparative studies examining the efficacy of these variants within PEP frameworks are limited.

  • PEPs and TCP Splitting: PEPs effectively split TCP connections to optimize performance across terrestrial and satellite segments. This can improve the overall transmission characteristics by handling queuing and retransmission in a more effective manner.

Study Focus

  1. Performance Metrics: The study assesses latency and throughput using a custom TCP-Splitting Satellite Emulation Framework (TSEF). The framework allows real-time measurement of traffic and the impact of AQM policies on TCP flow performance, specifically with respect to three traffic control approaches.

  2. Testing Methodology: Four testing scenarios were developed based on combinations of TCP flavors and AQM policies. The study utilized an extensive approach to mimicking real-world conditions in satellite environments, using full Linux implementations instead of mere simulations.

TCP Approaches Evaluated

  • Approach 1: CUBIC with classic drop-tail token bucket, acting as the control baseline.

  • Approach 2: CUBIC combined with CoDel AQM for improved latency management.

  • Approach 3: YeAH using CoDel AQM, focusing on low-latency transmissions.

  • Approach 4: BBR using a drop-tail token bucket, assessed primarily for its high-speed capabilities.

Study Findings and Insights

  • Settling Period: In the initial stages, YeAH outperformed others by reaching maximum throughput faster. BBR followed closely, with CUBIC lagging behind.

  • Stabilization Period: When subjected to transient conditions, CUBIC and BBR displayed efficient stabilization, sustaining throughput effectively during competition for bandwidth.

  • Aggregate Period: Notably, both CUBIC and BBR struggled with high latency during this phase. YeAH demonstrated favorable throughput with some fairness concerns.

  • Recovery Period: The ability of TCP variants to recover bandwidth after competition was measured. Traditional CUBIC was fastest, while YeAH took significantly longer to resume optimal throughput.

  • Isolated Period Performance: When isolated, YeAH maximized throughput and maintained low latency, making it a strong candidate for real-time applications; however, CUBIC exhibited notable efficiency, albeit with slightly higher latency in certain scenarios.

Conclusion

This study reiterates the importance of selecting appropriate TCP variants to enhance satellite network performance. Although YeAH showed exceptional promise in optimizing throughput and latency, CUBIC’s versatility and compatibility with AQM policies make it a robust choice for applications where fairness and speed are integral. The results point towards the necessity of fine-tuning TCP parameters and implementing effective AQM strategies to adapt to the high costs and unique challenges presented by satellite communications. Future research should explore the implementation of advanced AQM methods to further enhance fairness and performance across varying network conditions.