Whats the Best Way To Reduce Storage Requirements Without Losing Insights? Push AI To...
summary
ElastiFlow describes an OTel native edge processing node that runs an agentic AI (often gpt-oss-20b) and stores signals in local DuckDB storage, making routing decisions at the edge instead of forwarding everything. Alerts and correlated conditions of interest go to the observability platform; bulk logs stay local, aggressively aggregated or sampled. The setup halves storage and ingest costs, cuts mean time to insight, and uses an MCP endpoint to fetch raw edge data on demand (1 TB indexed cloud cost ~$1,800/month vs $15 local disk).
AI-written summary · about
chapters (14)
- 0:00Introduction
- 1:45Elastiflow’s network observability roots
- 3:09OpenTelemetry benefits
- 4:33Storage cost challenges
- 7:22Data filtering pyramid
- 9:48Correlating app and network traces
- 11:11Machine learning sampling
- 12:55Cost protection with ML
- 14:37Local storage safeguard
- 16:20Dynamic data retrieval
- 17:22Self-tuning observability
- 18:04Live demo: smart sampling
- 21:12Lessons learned and roadmap
- 22:10Q&A
related talks
05mEdge AI: Because the Cloud is Too Slow...
09mLightning Talk: Streaming and Processing Edge Vision Data in Real Time
13mNext-Gen Observability: reduce MTTR
10mWhy your observability is costing you too much?
51mSuper large, ultra fast, mega scaling cloud computing
18mThe Cutting Edge of AI and IoT
12mOptimizing AI Data Infrastructure with F5 and
24mWhat does AI mean for IoT development?