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

Abstract

With the deployment of mega constellations of Low-Earth-Orbit (LEO) satellites, low latency and high throughput Internet coverage is extended globally. Latency-sensitive applications can benefit from the inherent lower transmission delay of LEO satellite networks compared to traditional Geostationary-Earth-Orbit (GEO) satellite networks. Starlink employs a globally time-synchronized controller to manage the association of satellite-to-ground communication links with an interval of 15 seconds, at fixed 12-27-42-57 seconds of every minute. Latency spikes and packet losses can occur during the handover period which can degrade the performance of transport layer protocols including TCP and QUIC, which rely on similar congestion control algorithms for fair data transmission. In this paper, we investigate the impact of the frequent Starlink handover events on QUIC performance. By leveraging the predictable handover patterns to avoid unnecessary congestion window reduction, we improved the performance of QUIC by up to 35% in terms of completion time in both network emulation and real-world experiments over Starlink networks. Our approach is independent of specific loss-sensitive congestion control algorithms and can be easily generalized.