Skip to content

Monitoring dashboards

This branch vendors the Grafana dashboards needed for CRM observability under monitoring/dashboards/ so they can be imported directly into a Grafana instance backed by the repository's Prometheus datasource (uid prometheus). When adding a new Grafana dashboard you can either upload these JSON files directly or open one of the files in monitoring/dashboards/ and copy/paste its contents into Grafana's import dialog.

Grafana dashboards

File Title Purpose
dashboards/system-overview-grafana-dashboard.json CRM — System Overview Default operational view. Fast-loading (15 min window) health dashboard covering workload FPS, node CPU/memory, DRAM bandwidth, package/DRAM power, thermal headroom, IPC, UPI, PCIe, and observability status. Start here.
dashboards/fps-grafana-dashboard.json MXL FFmpeg FPS Application-level encoder/decoder FPS, speed ratio, and frame throughput.
dashboards/pcm-grafana-dashboard.json Intel PCM — k8s-w2 (FFmpeg-MXL Scenario 1) Concise CRM workload view for DRAM bandwidth, LLC behaviour, PCIe traffic, package power, and active frequency during scenario runs.
dashboards/pcm-extended-grafana-dashboard.json Intel PCM — Extended Hardware Counters (Deep Dive) Deep-dive investigation view. Detailed hardware counters: core/thread activity, IPC, L2/L3 cache, per-socket DRAM, power, frequency, PCIe, thermal headroom, and measurement interval.
dashboards/perfspect-grafana-dashboard.json Intel PerfSpect — k8s-w2 (FFmpeg-MXL Scenario 1) Top-down microarchitecture, IPC/CPI, NUMA, and latency analysis from the PerfSpect collector.

PMU is single-owner: Intel PCM and Intel PerfSpect cannot own the PMU simultaneously. Keep both dashboards installed if you want, but only the active collector will emit data.

LLC occupancy note: this repository's live Grafana dashboards use Prometheus-fed PCM/PerfSpect/FFmpeg metrics. Per-cgroup LLC occupancy is collected by the Intel RDT workflow and written into run results, so it is documented in the observability docs rather than queried live from the PCM dashboard JSON.

Importing a dashboard

  1. Open Grafana and navigate to Dashboards → Import.
  2. Either click Upload JSON file and select a file from monitoring/dashboards/, or open one of the JSON files there (for example dashboards/pcm-extended-grafana-dashboard.json) and copy/paste its contents into the import dialog.
  3. When Grafana prompts for a datasource, choose Prometheus with uid prometheus.
  4. Click Import.

The FPS dashboard includes a namespace variable that defaults to crm-workloads; if your encoder/decoder Pods run elsewhere, update that variable after import.

The System Overview dashboard also includes a namespace variable (same default) and an instance variable that is auto-populated from the PCM exporter targets.