Sustainability Research the Cloud Native Way
summary
Kepler (Kubernetes-based Efficient Power Level Exporter) uses eBPF and system counters to export energy and performance metrics to Prometheus, staying lightweight (~2% CPU, ~40MB). A small ML model maps counters to per-node, per-pod power; a demo lowers CPU frequency from ~4GHz to 2GHz to save ~30–40% node energy while the clever recommender plugs into Kubernetes VPA to double pod CPU requests (250→500 mCPU) and keep instructions-per-second stable. Default reaction time is about one minute and is configurable.
AI-written summary · about
chapters (8)
- 0:00Cloud native sustainability
- 2:06Metrics correlation and scheduling
- 4:13Kepler project for power
- 9:35Energy with Kepler and VPA
- 14:37Kepler deployment demo
- 23:16Workload testing with recommender
- 29:50Future directions and community
- 30:54Q&A on energy beyond CPU
related talks
20mMonitoring Energy Consumption and Building Energy Efficient Systems the Cloud Native Way
40mSustainability the Container Native Way
20mEnergy Efficient Placement of Edge Workloads
20mKepler - Optimized power usage monitoring for Kubernetes
42meBPF-Powered Observab...
28mExploring Kepler's Next Chapter: Achieving Cloud Native Sustainability With MCP Integration
05mProject Lightning Talk: Kepler Project Updates
12mKeynote: Cloud Native Sustainability - How You Can Help Right Now!