Network Fault Localization in a Complex Business System
To pin down the root cause of a network fault in a complex business system, engineers drove AVL Code with natural-language instructions — a full intelligent loop from 2.2GB of traffic captures to a fault-localization report.
Built with AVL Code + the Landi model
Overview
This is the re-diagnosis of a chronic network problem that had eluded direct localization for five years. The organization serves tens of millions of users across many internal business systems with complex traffic distribution; during trial operation, internet-facing services slowed to a crawl and a 10G isolation gateway nearly froze as internal-external traffic amplified almost 10x — yet repeated packet captures at the time never produced direct evidence of a loop. With AVL Code and the Landi model, and no extra tools to install or configure, the analysis parsed three traffic captures totaling 2.2GB and ran the full pipeline from traffic parsing to root-cause localization. The conclusion matched what was observed on site five years earlier — and surfaced several additional latent issues.
Key results
- Parsed 3 pcap captures totaling 2.2GB of traffic data
- Localized the key root cause: near-10x traffic amplification across the isolation gateway
- Generated a structured fault-localization report whose conclusions matched the on-site observations from 5 years earlier, surfacing several additional latent issues
- Completed the flow in about 15 minutes with no extra tooling — a 3–5x efficiency gain over manual analysis
Technical highlights
Practical value
Slashes troubleshooting cycles for network faults in complex business systems, helps ops teams quickly locate bottlenecks and security risks, and improves system stability and ops efficiency — while proving the real-world analysis capability of AVL Code on production-scale networks.
Artifacts
Session replays & reports are original records in Simplified Chinese · Built with AVL Code + the Landi model