SNX Fabric — Live Simulation with AI (graph neural network)snx.space ↗

loading topology…
SIMULATION UPDATE FOR 02 NOV 2026 Unsettled Updated daily · last 24 h · generated 2026-11-02 06:00 UTC · text
The simulation ran without interruption through the last 24 hours, advancing the orbital clock to day 26.1 of the 30-day window and logging 2,872 steps at roughly 30 seconds of wall time per 20-second step. Conditions across the fabric were steady: around 209 of the 2,488 pre-wired links were lit at any moment, with satellites handing customers and gateways between beams about 10,030 times over the day and 5,645 measured paths re-routing as a result.
Visibility on the measured paths was patchy in places: OneWeb1 → OneWeb4, StarcloudDC1 → gnd-starlink reachable 93 % and 59 % of the time (see the table), the rest close to continuous. The quickest crossing was StarcloudDC1 → LonestarDC1 at about 98 ms, the longest Kepler1 → Kepler12 at about 270 ms.
The AI path selector scored 2,872 more steps today and is now clearly ahead of OSPF, choosing paths that cost 0.2 ms of regret against OSPF's 10.1 ms; it agreed with OSPF's choice 87 % of the time and disagreed where it saw a link about to drop. Nearly every path it chose survived its full horizon (100.0 % against OSPF's 93.1 %). Its lead over OSPF shrank through the day.
Outlook: unsettled — pair reachability 95.5 %; the run continues and nothing needs intervention unless it persists.
—

Fabric

scroll to zoom · drag to pan · click a path tile to highlight
SNX node — SNXn-MEO backbone (two planes), SNXn-LEO edge (SSO) Kepler (AS 65001) — 16 sats, in-plane ISLs OneWeb (AS 65002) — 6 sats, bent-pipe Orbital data centers — StarcloudDC1 (AS 65101), LonestarDC1 (AS 65102) faded = unattached / not in phase / spare

Measured paths

AI path selector — live learning curve

A graph neural network (two graph-attention layers over the SNX fabric, running on a separate VM) picks a path for every SNX node pair at every 20 s simulation step, in the shadow of OSPF. Five steps later each pick is scored against the path that hindsight says was best, and the model trains on the result. Lower regret is better; survival is the share of picked paths whose links all stayed up through the horizon. Updated daily — snapshot of the last 24 hours, taken at … UTC.
AI selector (GNN) — regret, msOSPF — regret, msAI survival (right axis)

Run history — round-trip time

Run history — hop count