TCP's feedback-governed nature struggles with Starlink's high jitter, packet loss rate, and regular latency jumps. Traditional TCP algorithms, like TCP Reno, perform poorly due to their sensitivity to packet loss and slow recovery. BBR or CUBIC Algorithm performs better by estimating the path bandwidth and adjusting the sending rate based on queue formation onset, but requires careful implementation to handle Starlink’s latency variability.
Future candidates for LEO transmission
- BBR: Deferring rate inflation during satellite handovers could optimize performance.
- SACK: Helps distinguish between isolated packet loss and congestion-induced loss.
- ECN: Could provide clearer signals for congestion control by distinguishing between queue formation and transient events like satellite handovers.