In the high-stakes environment of Low Earth Orbit (LEO) satellite networking, the “best” route is often a moving target. During the AIORI-2 Hackathon, team ByteCoders from Sharda University discovered that traditional, reactive routing simply can’t keep up with satellites moving at 27,000 km/h. By implementing the proactive architecture proposed in RFC 9717, we proved that a “slower” path on paper is often the fastest path in reality if it actually stays connected.
1. The LEO Challenge: Geometry in Motion
Standard routing assumes nodes are stationary. In a LEO constellation, the topology changes every second. Our baseline Model 1 (CGR), while mathematically “perfect,” often selected paths that would vanish mid-packet.
To solve this, we pivoted to RFC 9717, which suggests a “Path Computation Element” (PCE) that uses orbital schedules to predict the future. We compared this against a decentralized ad-hoc protocol (DSDV) to see which could handle the chaos of space better.
2. A 3-Tiered Routing Simulation
Our framework was built on a shared “physics engine” using the Skyfield library to ensure that all three models faced the exact same orbital dynamics.
The Model Comparison:
- Model 1 (The Benchmark): A centralized Dijkstra implementation. It finds the shortest path now, regardless of the future.
- Model 2 (Decentralized): A simply-based simulation of DSDV. Nodes “talk” to neighbors to discover the map.
- Model 3 (The RFC 9717 Solution): A proactive PCE that checks if a path will still exist 30 seconds into the future before authorizing it.
3. Implementation Stack
We deployed our simulation on the AIORI Testbed, utilizing high-compute nodes to handle the KDTree graph optimizations required for thousands of potential satellite links.
| Component | Technology | Role |
|---|---|---|
| Physics Engine | Skyfield & Python 3.11 | Orbital propagation and TLE processing. |
| Graph Logic | NetworkX & SciPy | Building the KDTree for neighbor discovery. |
| Event Simulation | Simpy | Modeling decentralized message passing. |
| Optimization | Dijkstra’s Algorithm | Centralized pathfinding benchmark. |
4. Key Findings: The “Price of Stability”
Our simulation produced a definitive validation for the RFC 9717 architecture. While Model 1 found a path with 42.5ms latency, that path was flagged as “BRITTLE.”
- Model 3 (Proactive) rejected the 42.5ms path and chose a 46.8ms path instead.
- The Result: The 42.5ms path failed after only 4 seconds. The 46.8ms path remained stable for the entire 30-second window.
- Conclusion: A 4.3ms “latency tax” is a small price to pay for a connection that doesn’t drop.
5. Lessons from the Stars
Our sprints proved that in dynamic networks, Predictive Intelligence > Reactive Speed. 1. Centralization Wins: In LEO, decentralized protocols like DSDV converge too slowly (~25s) compared to the rate of topological change.
- Standardization Matters: RFC 9717 provided the architectural blueprint that moved our project from a simple “shortest path” script to a resilient, space-grade routing framework.
“Working with RFC 9717 transformed our project from a simple algorithm to a ‘smart’ predictive system.” — Roshita Verma, Team Lead
6. Open Source Contribution
We have open-sourced our Contact Graph Routing (CGR) module. It serves as a modular, reusable Dijkstra benchmark for any researcher looking to test satellite or delay-tolerant network (DTN) routing.
Read the full report
