1 hour ago
14
When your internet router sits in your living room, it stays put. When it’s a satellite screaming across the sky at orbital velocity, the rules change. Cloudflare’s CTO Dane Knecht went public with a proposal for SpaceX’s Starlink team to co-develop congestion-control algorithms purpose-built for low-Earth orbit satellite networks, a technical challenge that’s becoming harder to ignore as Starlink’s user base balloons.
Elon Musk reportedly passed the message along to the Starlink team, though no formal partnership has materialized.
The problem with treating satellites like fiber
Most of the internet runs on congestion-control algorithms like CUBIC, a protocol that manages how much data gets pushed through a connection at any given moment. CUBIC works well when conditions are predictable: stable routes, consistent latency, minimal packet loss. It was designed for terrestrial networks where the path between two points doesn’t fundamentally change every few minutes.
LEO satellite networks break those assumptions in several ways. Starlink satellites orbit at roughly 550 kilometers above Earth, completing a full lap every 90 minutes or so. As a user’s connection hands off from one satellite to another, the route data takes can shift abruptly. Round-trip times, the delay between sending a packet and receiving confirmation it arrived, can spike without warning.
For algorithms like CUBIC, those spikes look indistinguishable from network congestion. The protocol’s natural response is to throttle back, reducing throughput to ease what it perceives as a traffic jam. But there’s no jam. The satellite just moved. The result is slower speeds and higher latency for users, even when the network has plenty of capacity.
Academic and industry research published through 2025 highlighted these performance trade-offs in detail, documenting how standard congestion-control mechanisms consistently underperform on Starlink links compared to traditional broadband.
Why this matters now
Starlink hit approximately 12 million subscribers by the end of Q2 2026, doubling its total from the year prior. Traffic from Starlink passing through Cloudflare’s network rose by 2.3 times during 2025.
Knecht’s proposal isn’t coming out of nowhere. Cloudflare has extensive experience tuning TCP and QUIC performance across varied network conditions, particularly lossy or unstable connections. The company has also worked with Starlink previously on infrastructure improvements aimed at boosting performance.
What a LEO-native algorithm would look like
The core challenge is teaching congestion-control systems to distinguish between real congestion and the inherent variability of satellite networks. A LEO-aware algorithm would need to account for predictable handover events, where a connection transfers from one satellite to the next, without interpreting the resulting latency spike as a signal to slow down.
It would also need to handle the fact that routes through a satellite constellation aren’t static. Data might travel through three satellites on one pass and five on the next, with each hop adding variable delay.
QUIC, the transport protocol that now carries a significant share of web traffic, offers some advantages here. Unlike TCP, which is implemented in operating system kernels and difficult to update, QUIC runs in user space. That makes it far easier to experiment with and deploy new congestion-control strategies without waiting for OS-level patches to propagate across billions of devices.
Cloudflare has been a major contributor to QUIC adoption and optimization, which gives it a natural platform for testing LEO-specific approaches at scale.
The competitive landscape for satellite internet
Starlink isn’t operating in a vacuum, at least not competitively. Amazon’s Project Kuiper is preparing its own LEO constellation, and several other ventures are in various stages of deployment.
The absence of a formal partnership announcement means this remains a public pitch rather than a confirmed project.
Disclosure: This article was edited by Editorial Team. For more information on how we create and review content, see our Editorial Policy.
English (US) ·