Scale your business in real-time using ML in Microsoft Fabric Real-time Analytics
summary
Custo (Kusto) uses native time‑series functions like series_decompose and series_forecast to predict CPU demand from three weeks of history, combine ingestion and query drivers, and run a hybrid predictive+reactive autoscaler. The deployed system autosizes 10,000 clusters across ~40 regions on live telemetry (serving ~251 PB and 3.4B queries), claims roughly 50% cost reduction versus a reactive-only policy, enforces min/max node caps, and supports custom Python/R models when native forecasts aren’t enough.
AI-written summary · about
chapters (6)
- 0:00Scaling with ML in Fabric
- 7:06Kusto real-time capabilities
- 15:34Reactive vs predictive scaling
- 20:16Time series analysis demo
- 34:41Real-time autoscaling in production
- 45:34Q&A
related talks
14mPerformance Optimization in High-Volume Transaction Systems
07mScaling Predictive Manufacturing Analytics: Kubernetes-Powered CRM
30mReal-Time ML services: From ideation to production
42mReimagine Real-Time Intelligence with Microsoft Fabric
09mReal-time Intelligence in Microsoft Fabric
26mReal Time analytics - Sense analyze generate insights with Fabric
13mBuilding Scalable AI/ML Platforms for Industrial IoT
31mFailing to Autoscale?