MirrorNet: High-fidelity and Scalable Network Emulation for Software-defined WAN
summary
MirrorNet recreates a production SD‑WAN by mirroring controller, data plane, and traffic and claims second‑by‑second fidelity while cutting storage dramatically (1300 GB/day per plane reduced by storing ~0.003% of changes). Key techniques are incremental storage and replay, dependency‑preserving network updates, and a custom orchestrator that deploys multi‑host GNS3 emulations. Deployed in TWAN for 2+ years, it enables parallel failure injection, fine‑tuning of TE parameters, and provisioning hundreds of times faster than OpenStack (≈8× faster than single‑host GNS3).
AI-written summary · about
chapters (20)
- 0:00MirrorNet introduction
- 0:19Why MirrorNet is needed
- 1:01Production network challenges
- 2:46Untraceable network failures
- 4:50Parameter tuning difficulties
- 6:36Single-plane testing limits
- 8:02Accurate capacity assessment
- 8:45MirrorNet design goals
- 9:28High-fidelity emulation techniques
- 10:32Network state storage
- 11:35Incremental replay method
- 13:01Consistent network updates
- 13:45Dependency-preserving injection
- 14:27Tunnel mapping for traffic
- 15:31Emulation scalability challenges
- 15:52Non-blocked optimization
- 16:34Implementation details
- 17:16Troubleshooting benefits
- 18:17Operational lessons learned
- 20:24Limitations and future work
related talks
22mTransparent MultiNIC Routing for Large AI Models
23mDIMINET: On Demand Multi-site Connectivity for Edge Infrastructures
16mOneWAN is better than two: Unifying a split WAN architecture
16mEnhancing Global Network Monitoring with Magnifier
10mLightning Talk: NSM, Network Service Mesh at Scale for Telco Networking
17mMeta’s Network Journey to Enable AI
24mMetron: NFV Service Chains at the True Speed of the Underlying Hardware
16mStarryNet: Empowering Researchers to Evaluate Futuristic Integrated Space and Terrestrial