O11y in the Evolution of AI-Era Networking Infrastructure
summary
Claims concrete upgrades: cites OpenTelemetry and tools like fiddler AI and erase AI for model drift and explainability, describes silicon-level telemetry and 800 gig port speed targets and a 51.2 Tb/s per-box processing figure, outlines a working self-healing pipeline that reroutes traffic and restarts services automatically, lists metrics used (latency, throughput, model accuracy, GPU power/temperature), and shows digital twin simulations and distributed tracing actually used to reproduce and fix failures in real deployments.
AI-written summary · about
chapters (8)
- 0:00Introduction to Observability in AI
- 0:36Historical Evolution of Network Observability
- 2:08Modern Cloud and AI Era Tools
- 3:07Importance of Network Observability in AI
- 5:39Challenges in Network Observability
- 10:25Advanced Telemetry and Self-Healing Networks
- 16:49Future of Observability in AI Networks
- 19:12Conclusion
related talks
10mApproach to AI at a large scale
26mAI Traffic Is Different: Rethinking Proxy Arc...
05mOpening Remarks
40mOptimizing AI Applications for Production with Observability
20mAI Impact on the Backbone
24mWhat Do You Mean It’s Broken? When AI Meets the Harsh Reality of Te...
09mAI for NI
14mImplementing AI-Powered O11y with Splunk