LDMS Metrics¶
This page catalogs the metrics collected by the Lightweight Distributed Metric Service (LDMS) samplers running on compute nodes. LDMS provides high-frequency, low-overhead OS-level metrics.
Collection method¶
| Property | Value |
|---|---|
| Agent | LDMS sampler daemon (ldmsd) on each compute node |
| Aggregation | LDMS aggregator daemon on the telemetry entry node |
| Default interval | 10 seconds (configurable via ldms_sample_interval in telemetry_config.yml) |
| Kafka topic | ldms_telemetry (configurable via kafka_topic_ldms) |
| Storage | VictoriaMetrics time-series database |
meminfo sampler¶
Collects data from /proc/meminfo. Provides memory utilization metrics.
| Metric | Unit | Description |
|---|---|---|
ldms_meminfo_MemTotal |
kB | Total physical memory available. |
ldms_meminfo_MemFree |
kB | Free memory (not used by any process or cache). |
ldms_meminfo_MemAvailable |
kB | Estimated memory available for starting new applications. |
ldms_meminfo_Buffers |
kB | Memory used by kernel buffers. |
ldms_meminfo_Cached |
kB | Memory used by the page cache (file data). |
ldms_meminfo_SwapTotal |
kB | Total swap space. |
ldms_meminfo_SwapFree |
kB | Free swap space. |
ldms_meminfo_Dirty |
kB | Memory waiting to be written to disk. |
ldms_meminfo_Writeback |
kB | Memory actively being written to disk. |
ldms_meminfo_AnonPages |
kB | Memory used by anonymous (non-file-backed) pages. |
ldms_meminfo_Slab |
kB | In-kernel data structures cache. |
ldms_meminfo_HugePages_Total |
Count | Total number of hugepages allocated. |
ldms_meminfo_HugePages_Free |
Count | Number of unused hugepages. |
vmstat sampler¶
Collects data from /proc/vmstat. Provides virtual memory statistics.
| Metric | Unit | Description |
|---|---|---|
ldms_vmstat_pgpgin |
Pages | Total pages paged in from disk (cumulative). |
ldms_vmstat_pgpgout |
Pages | Total pages paged out to disk (cumulative). |
ldms_vmstat_pswpin |
Pages | Pages swapped in (cumulative). |
ldms_vmstat_pswpout |
Pages | Pages swapped out (cumulative). |
ldms_vmstat_pgfault |
Count | Total page faults (minor + major, cumulative). |
ldms_vmstat_pgmajfault |
Count | Major page faults (required disk I/O, cumulative). |
ldms_vmstat_oom_kill |
Count | Number of OOM (Out-Of-Memory) kill events. |
ldms_vmstat_numa_hit |
Count | Memory allocations satisfied from the preferred NUMA node. |
ldms_vmstat_numa_miss |
Count | Memory allocations that fell back to a remote NUMA node. |
procstat sampler¶
Collects data from /proc/stat. Provides CPU utilization metrics.
| Metric | Unit | Description |
|---|---|---|
ldms_procstat_user |
Jiffies | Time spent in user mode (cumulative, per CPU and total). |
ldms_procstat_nice |
Jiffies | Time spent in user mode with low priority (nice). |
ldms_procstat_system |
Jiffies | Time spent in kernel mode. |
ldms_procstat_idle |
Jiffies | Time spent idle. |
ldms_procstat_iowait |
Jiffies | Time waiting for I/O to complete. |
ldms_procstat_irq |
Jiffies | Time servicing hardware interrupts. |
ldms_procstat_softirq |
Jiffies | Time servicing software interrupts. |
ldms_procstat_steal |
Jiffies | Time stolen by hypervisor (VM environments). |
ldms_procstat_processes |
Count | Number of processes created since boot (cumulative). |
ldms_procstat_procs_running |
Count | Number of processes currently in runnable state. |
ldms_procstat_procs_blocked |
Count | Number of processes currently blocked on I/O. |
ldms_procstat_context_switches |
Count | Total context switches since boot (cumulative). |
procnetdev sampler¶
Collects data from /proc/net/dev. Provides per-interface network
statistics.
| Metric | Unit | Description |
|---|---|---|
ldms_procnetdev_rx_bytes |
Bytes | Total bytes received on the interface (cumulative). |
ldms_procnetdev_rx_packets |
Count | Total packets received (cumulative). |
ldms_procnetdev_rx_errors |
Count | Total receive errors (cumulative). |
ldms_procnetdev_rx_dropped |
Count | Total received packets dropped (cumulative). |
ldms_procnetdev_tx_bytes |
Bytes | Total bytes transmitted on the interface (cumulative). |
ldms_procnetdev_tx_packets |
Count | Total packets transmitted (cumulative). |
ldms_procnetdev_tx_errors |
Count | Total transmit errors (cumulative). |
ldms_procnetdev_tx_dropped |
Count | Total transmitted packets dropped (cumulative). |
Metric labels¶
All LDMS metrics include the following common labels:
| Label | Description |
|---|---|
host |
Hostname of the compute node. |
component_id |
Unique component identifier assigned by LDMS. |
job_id |
Slurm job ID (if job-level tagging is enabled). |
interface |
Network interface name (procnetdev metrics only). |
cpu |
CPU core number (procstat per-core metrics only). |
Info
- Telemetry Config -- LDMS configuration parameters.
- Idrac Metrics -- Hardware-level metrics from iDRAC.
- DCGM Metrics -- In-band GPU telemetry metrics (DCGM/ROCm).