Submitted: March 2025
Abstract
Low-Earth orbit (LEO) satellite constellations gain an ever-increasing relevancy when it comes to providing low-latency, high-bandwidth Internet connectivity around the globe. The highly dynamic nature of these networks, caused by the enormous speeds at which satellites move relative to the Earth’s surface, makes for a challenging environment that differs significantly from the traditional terrestrial Internet and its stationary infrastructure. A single satellite covers only a small and shifting fraction of the Earth’s surface, leading to limited visibility times from a ground user’s perspective and highly volatile end-to-end connectivity patterns with frequently changing routes between two points on Earth. With this thesis, we aim to provide an understanding of the influence of route changes on well-established transport layer technology, particularly congestion control algorithms (CCAs), and utilize the gained insights to devise CCAs tailored for deployment in LEO satellite networks.
We developed a comprehensive model that approximatively describes the impact of route changes on flows controlled by the loss-based CCAs Reno and CUBIC. The model further distinguishes between the transport layer protocols TCP and QUIC and the implications of route changes on their loss detection algorithms. Based on the derived route change impact factors, we devised OrbitalReno and OrbitalCUBIC, orbit-ready variants of Reno and CUBIC that adjust their state variables appropriately in the presence of route changes. We implemented the orbital CCAs in QUIC and extended them to prevent spurious loss events in QUIC.
We evaluated the performance of the orbital CCAs over a virtual end-to-end link that emulated the real-world LEO constellation Starlink. Regarding transmitted data volume, we observed performance enhancements of senders that employ an orbital CCA by up to 60 % compared to its standard counterpart. Furthermore, our results show that the orbital CCAs provide a viable alternative to route selection algorithms focusing on route stability and longevity. While these routing algorithms help the standard CCAs to overcome the issue of goodput degradation through route changes at the expense of higher delays, the orbital CCAs facilitate both high goodput and low delays. The performance enhancements could be even more significant if the orbital CCAs are supplemented by techniques to overcome spurious retransmissions.