Energy Efficient Placement of Edge Workloads
Talkat Kubernetes on Edge Day North America 2022 · on Tuesday, 2022-10-25· at 15:45· Room 250 ABC, Huntington Place: 1 Washington Blvd·
summary
Kepler uses eBPF to probe kernel counters and exports per-pod energy metrics (CPU, GPU, DRAM) as Prometheus time series with low agent overhead, and a Kepler model server supplies linear regression models (CPU and DRAM) with online learning when hardware meters are unavailable. A three-node OpenShift demo labels nodes green/yellow/red from a carbon-intensity forecaster, applies taints and 5s tolerations to migrate long-running workloads, and exposes difficulties: sparse public carbon APIs and missing time points.
AI-written summary · about
chapters (7)
- 0:00Start
- 1:39Background
- 5:41Kepler Model Server's Models
- 6:28Model Server and Kepler
- 7:58Use Cases
- 9:25Carbon Intensity Forecaster
- 18:29Road Ahead
related talks
20mMonitoring Energy Consumption and Building Energy Efficient Systems the Cloud Native Way
20mKepler - Optimized power usage monitoring for Kubernetes
35mSustainability Research the Cloud Native Way
28mExploring Kepler's Next Chapter: Achieving Cloud Native Sustainability With MCP Integration
05mProject Lightning Talk: Kepler Project Updates
18mKeynote: Sustainability Chronicles: Innovate Through Green Technology with...
42meBPF-Powered Observab...
26mAccelerating Science, Decelerating Carbon: Sustainable Computing...