Post by Thomas Harvey

A network that cannot read its own state cannot act on its own state. The fabric was healthy. IS-IS converged. BGP sessions established. 17 health checks passed. The diamond network was running. But if a session dropped and recovered at 2am, nothing knew. If latency crept past threshold, nothing noticed. Installment 8 of The Thinking Network documents the shift from reactive SNMP polling to high-frequency streaming telemetry. Using gNMI on Nokia SR Linux, the fabric now continuously streams its operational state directly to Prometheus. The first run returned flat 0.0ms L3 latency across every cycle. Not a win. A symptom. The sensing loop was clean but the client container had no IPv4 address and no route to the gateway. It was measuring nothing and reporting it confidently. Two commands fixed it. Real numbers came back. L2 EVPN: 0.056ms L3 VPRN: 2.06ms BGP sessions: 3 of 3 established The distinction that matters: a monitored network is watched from the outside. An aware network has sensing embedded in its own operational layer. Its state becomes legible to the systems that will eventually act on it. This is the data ingestion pipeline that predictive AI modeling requires. Next installment: the actuator reads this data, detects a breach, and reaches into BGP. The network does something on its own for the first time. Full breakdown with architecture, fixes, and logs at the link. https://lnkd.in/g5ikHvKf #NetworkEngineering #Automation #SRLinux #Telemetry #AutonomousNetworks