UFM Metrics¶
This page catalogs the InfiniBand fabric metrics collected by the NVIDIA Unified Fabric Manager (UFM) Prometheus exporter. These metrics are scraped by vmagent and stored in VictoriaMetrics.
Collection Method¶
| Property | Value |
|---|---|
| Collection tool | UFM Prometheus Exporter |
| Protocol | Prometheus scrape over HTTPS |
| Default port | 9001 |
| Default interval | 30 seconds (configurable via scrape_interval in telemetry_config.yml) |
| Storage | VictoriaMetrics time-series database |
UFM Telemetry Metrics Endpoint¶
Endpoint Overview¶
| Property | Detail |
|---|---|
| URL | https://<ufm-telemetry-host>:9001/metrics |
| Format | Prometheus text exposition over HTTPS |
| Purpose | Exposes InfiniBand fabric telemetry counters across all monitored ports |
Sample Metrics¶
| Metric | Description |
|---|---|
PortXmitDataExtended |
Cumulative count of bytes sent |
PortRcvDataExtended |
Cumulative count of bytes received |
PortRcvErrorsExtended |
Cumulative count of inbound error packets |
PortXmitDiscardExtended |
Cumulative count of outbound packets discarded |
phy_state |
Current physical state of the port |
link_speed_active |
Presently active link speed |
temperature |
Module or ASIC temperature reading (°C) |
Example: link_speed_active metric visualization
References¶
- UFM Telemetry Manager Plugin - UFM Telemetry Manager (UTM) Plugin documentation
- NVIDIA UFM Enterprise - Telemetry - Telemetry configuration and usage
Info
- Telemetry Config -- UFM telemetry configuration parameters.
- NVIDIA UFM Enterprise User Manual - UFM documentation.
- Idrac Metrics -- Hardware-level metrics from iDRAC.
- Ldms Metrics -- OS-level metrics from LDMS.