eBPF-Powered Observab...
summary
eBPF hooks collect CPU/GPU power counters (RAPL, NVML, ACPI) and expose energy in joules per node and per pod via Kepler as Prometheus metrics, then Grafana/alerts handle thresholds; idle power is attributed proportionally across containers so optimization targets are identifiable. A prototype pipeline sends eBPF features through Pub/Sub/Dataflow to BigQuery and uses TimesFM zero‑shot forecasting detects anomalies for realtime network anomaly detection, claiming low overhead compared with sidecar telemetry and Kepler + eBPF exports joules per pod to Prometheus.
AI-written summary · about
chapters (7)
- 0:00eBPF observability intro
- 3:14Sustainable computing challenges
- 7:12What is eBPF
- 16:05eBPF for sustainable computing
- 24:50Network anomaly detection
- 32:40Key takeaways
- 33:45Q&A
related talks
17mKepler: Powering up the efficiency of your cluster
35mSustainability Research the Cloud Native Way
20mMonitoring Energy Consumption and Building Energy Efficient Systems the Cloud Native Way
20mKepler - Optimized power usage monitoring for Kubernetes
18mFind out what your processes are capable of with ebpf
20mEnergy Efficient Placement of Edge Workloads
40mSustainability the Container Native Way
21mAuto-Instrumentation for GPU Performance using eBPF