MetricsDescription.md

# Metrics Description

## CPU (perf & turbostat-derived)

### Perf

Perf:Branches.*[count] — Number of branch instructions executed (cumulative counter).

Perf:Branch_Misses.*[count] — Number of branch prediction misses (cumulative counter).

Perf:Cycles.*[count] — CPU cycles executed (cumulative counter).

Perf:Instructions.*[count] — Instructions retired (cumulative counter).

Perf:IPC.*[-] — Instructions per cycle (IPC). Typically computed or sampled (instantaneous / derived).

Perf:BranchUtilization.*[%] — Branch unit utilization percent (sampled/instant).

Perf:CacheUtilization.*[%] — Cache utilization percent (sampled/instant).

Perf:StallRatio.*[%] — Fraction of cycles spent stalled (sampled/instant).

Perf:OccupancyRatio.*[%] — Execution-unit occupancy/saturation percent (sampled/instant).

### Turbostat (per-core / package readings and derived)

Fields with \\<n\\> repeat per CPU core index. Fields with .- indicate package-level or special entries present in headers.

#### Avg_MHz (per-core)

Turbostat:Avg_MHz.\<n\>.*[MHz] — Average frequency of core \<n\> in the sample interval. (instant / sampled)

Turbostat:Avg_MHz.-.*[MHz] — Package-level / special average frequency entry. (instant)

#### Busy% (per-core)

Turbostat:Busy%.\<n\>.*[%] — Percent of sample interval core \<n\> was busy. (sampled)

Turbostat:Busy%.-.*[%] — Package-level / special Busy% entry. (sampled)

#### Bzy_MHz (per-core)

Turbostat:Bzy_MHz.\<n\>.*[MHz] — Effective frequency when core \<n\> was busy (MHz). (sampled / derived)

Turbostat:Bzy_MHz.-.*[MHz] — Aggregate / special entry. (sampled)

#### TSC_MHz (per-core)

Turbostat:TSC_MHz.\<n\>.*[MHz] — Timestamp Counter frequency estimate for core \<n\>. (sampled / diagnostic)

Turbostat:TSC_MHz.-.*[MHz] — Aggregate / special entry.

#### IRQ (per-core)

Turbostat:IRQ.\<n\>.*[count] — Number of interrupts handled by core \<n\> (cumulative counter).

Turbostat:IRQ.-.*[count] — Aggregate / special.

#### SMI (per-core)

Turbostat:SMI.\<n\>.*[count] — System Management Interrupts on core \<n\> (cumulative counter).

Turbostat:SMI.-.*[count] — Aggregate / special.

#### CPU C-state residency (per-core)

Turbostat:CPU%c1.\<n\>.*[%] — Percent time core \<n\> spent in C1 low-power state (sampled).

Turbostat:CPU%c1.-.*[%] — Aggregate / special.

Turbostat:CPU%c6.\<n\>.*[%] — Percent time core \<n\> spent in C6 deep low-power state (sampled).

Turbostat:CPU%c6.-.*[%] — Aggregate / special.

#### Core & Package temps / power / package state / utilization

Turbostat:CoreTmp.\<n\>.*[C] — Temperature of core \<n\> (°C). (sampled)

Turbostat:CoreTmp.-.*[C] — Aggregate / special core temperature. (sampled)

Turbostat:PkgTmp.\<n\>.*[C] — Package temperature reading (may appear for multiple package indices). (sampled)

Turbostat:PkgTmp.-.*[C] — Aggregate package temperature. (sampled)

Turbostat:Pkg%pc2.\<n\>.*[%] — Package time fraction spent in PC2 power state (percent). (sampled)

Turbostat:Pkg%pc6.\<n\>.*[%] — Package time fraction spent in PC6 power state (percent). (sampled)

Turbostat:PkgWatt.\<n\>.*[W] — Package-level power consumption (watts). (sampled)

Turbostat:RAMWatt.\<n\>.*[W] — RAM subsystem power (watts). (sampled)

Turbostat:PKG_%.\<n\>.*[%] — Package utilization percent (sampled/derived).

Turbostat:RAM_%.\<n\>.*[%] — RAM utilization percent (sampled/derived).

Note: header sometimes lists specific core indices (e.g., .0, .24), \<n\> generalizes these. Replace \<n\> with actual core index when referring to concrete columns.

#### CPU utilization (per core & average)

CPU:CPU\<n\>Utilization.*[%] — Utilization percent of CPU core \<n\> (sampled / instant).

AverageCPUUtilization.*[%] — Average CPU utilization across tracked cores (sampled / derived).

## Memory (perf-derived & /proc/meminfo style)

### Perf-derived memory metrics

Perf:Cache_Misses.*[count] — Cache misses (cumulative counter).

Perf:Cache_References.*[count] — Cache reference accesses (cumulative counter).

Perf:Cycle_activity_cycles_l3_miss.*[count] — Cycles lost due to L3 cache misses (cumulative counter).

Perf:Cycle_activity_stalls_l3_miss.*[count] — Stalls caused by L3 misses (cumulative counter).

Perf:Hle_retired_aborted.*[count] — HLE (hardware lock elision) aborted transactions (cumulative counter).

Perf:Hle_retired_commit.*[count] — HLE committed transactions (cumulative counter).

Perf:Mem_trans_retired_load_latency_gt_128.*[count] — Loads with latency \>128 cycles (cumulative counter).

Perf:Mem_trans_retired_load_latency_gt_256.*[count] — Loads with latency \>256 cycles (cumulative counter).

Perf:Mem-loads.*[count] — Memory load operations (cumulative counter).

Perf:Mem-stores.*[count] — Memory store operations (cumulative counter).

### Memory snapshot / /proc/meminfo-style fields (units in kB, typically instantaneous/sample):

Memory:Active.*[kB] — Amount of active (recently used) memory. (instant)

Memory:AnonHugePages.*[kB] — Anonymous hugepages memory. (instant)

Memory:AnonPages.*[kB] — Anonymous pages (non-file-backed). (instant)

Memory:Bounce.*[kB] — Bounce buffer memory used (DMA-related). (instant)

Memory:Buffers.*[kB] — Kernel buffers. (instant)

Memory:Cached.*[kB] — File pages cached in memory. (instant)

Memory:CommitLimit.*[kB] — Commit limit for virtual memory. (instant)

Memory:Committed_AS.*[kB] — Committed virtual memory. (instant)

Memory:DirectMap1G.*[kB] — 1 GiB direct-mapped pages. (instant)

Memory:DirectMap2M.*[kB] — 2 MiB direct-mapped pages. (instant)

Memory:DirectMap4k.*[kB] — 4 KiB direct-mapped pages. (instant)

Memory:Dirty.*[kB] — Pages waiting to be written to storage. (instant)

Memory:HardwareCorrupted.*[kB] — Memory marked as hardware-corrupted. (instant)

Memory:Hugepagesize.*[kB] — Hugepage size. (instant)

Memory:Hugetlb.*[kB] — hugetlb allocations. (instant)

Memory:Inactive.*[kB] — Inactive page cache. (instant)

Memory:KReclaimable.*[kB] — Kernel reclaimable memory. (instant)

Memory:KernelStack.*[kB] — Kernel stack usage. (instant)

Memory:Mapped.*[kB] — Mapped files/devices. (instant)

Memory:MemAvailable.*[kB] — Estimated available memory. (instant)

Memory:MemFree.*[kB] — Free physical memory. (instant)

Memory:MemTotal.*[kB] — Total physical memory. (instant)

Memory:Mlocked.*[kB] — Memory locked by mlock. (instant)

Memory:NFS_Unstable.*[kB] — NFS unstable writeback memory. (instant)

Memory:PageTables.*[kB] — Memory used by page tables. (instant)

Memory:SReclaimable.*[kB] — Reclaimable slab memory. (instant)

Memory:SUnreclaim.*[kB] — Unreclaimable slab memory. (instant)

Memory:Shmem.*[kB] — Shared memory usage. (instant)

Memory:ShmemHugePages.*[kB] — Shared memory hugepages. (instant)

Memory:ShmemPmdMapped.*[kB] — PMD-mapped shared memory. (instant)

Memory:Slab.*[kB] — Kernel slab usage. (instant)

Memory:SwapCached.*[kB] — Swap cached memory. (instant)

Memory:SwapFree.*[kB] — Free swap space. (instant)

Memory:SwapTotal.*[kB] — Total swap space. (instant)

Memory:Unevictable.*[kB] — Unevictable pages. (instant)

Memory:VmallocChunk.*[kB] — Largest contiguous vmalloc chunk. (instant)

Memory:VmallocTotal.*[kB] — Total vmalloc address space. (instant)

Memory:VmallocUsed.*[kB] — Vmalloc used. (instant)

Memory:Writeback.*[kB] — Currently writing back pages. (instant)

Memory:WritebackTmp.*[kB] — Temporary writeback usage. (instant)

Memory:MemoryUtilization.*[%] — Memory utilization percent (sampled/derived).

## Storage (Storage: prefix)

Storage:reads_completed_successfully.*[count] — Number of successful read operations (cumulative counter).

Storage:reads_merged.*[count] — Number of merged read requests (cumulative counter).

Storage:sectors_read.*[sectors] — Sectors read (cumulative counter). Multiply by sector size to get bytes.

Storage:writes_completed.*[count] — Number of successful write operations (cumulative counter).

Storage:writes_merged.*[count] — Number of merged write requests (cumulative counter).

Storage:sectors_written.*[sectors] — Sectors written (cumulative counter).

Storage:discards_completed_successfully.*[count] — Successful discard (trim) operations (cumulative counter).

Storage:discards_merged.*[count] — Merged discard requests (cumulative counter).

Storage:sectors_discarded.*[sectors] — Discarded sectors (cumulative).

Storage:time_spent_reading_(ms).*[ms] — Total milliseconds spent on reads in sample window (sampled/aggregate).

Storage:time_spent_writing_(ms).*[ms] — Total milliseconds spent on writes in sample window (sampled/aggregate).

Storage:I/Os_currently_in_progress.*[count] — Current number of I/O operations in progress (instant).

Storage:time_spent_doing_I/Os_(ms).*[ms] — Total milliseconds spent doing I/Os (sampled/aggregate).

Storage:weighted_time_spent_doing_I/Os_(ms).*[ms] — Weighted I/O time metric (sampled/aggregate).

Storage:time_spent_discarding.*[ms] — Milliseconds spent processing discard operations (sampled).

## GPU (device-specific fields; device index replaced by \<n\>)

### Device-specific fields

GPU: pcie.link.gen.current.\<n\>.*[generation] — Current PCIe link generation between GPU \<n\> and host (categorical).

GPU: pcie.link.width.current.\<n\>.*[lanes] — Current PCIe link width (lanes).

GPU: accounting.buffer_size.\<n\>.*[MB] — Accounting buffer size used for GPU memory accounting (MB). (instant)

GPU: fan.speed.\<n\>.*[%] — Fan speed percent or relative metric for GPU \<n\> (sampled).

GPU: memory.used.\<n\>.*[MiB] — GPU memory used (MiB) (instant).

GPU: memory.free.\<n\>.*[MiB] — GPU memory free (MiB) (instant).

GPU: utilization.gpu.\<n\>.*[%] — GPU core utilization percent. (sampled)

GPU: utilization.memory.\<n\>.*[%] — GPU memory subsystem utilization percent. (sampled)

GPU: utilization.encoder.\<n\>.*[%] — Hardware encoder utilization. (sampled)

GPU: utilization.decoder.\<n\>.*[%] — Hardware decoder utilization. (sampled)

GPU: utilization.jpeg.\<n\>.*[%] — JPEG unit utilization. (sampled)

GPU: utilization.ofa.\<n\>.*[%] — OFA unit utilization (if present). (sampled)

GPU: encoder.stats.sessionCount.\<n\>.*[count] — Number of encoder sessions (cumulative or sampled, depending on collector).

GPU: encoder.stats.averageFps.\<n\>.*[fps] — Average encoder FPS (sampled / aggregated).

GPU: encoder.stats.averageLatency.\<n\>.*[ms] — Average encoder latency (ms) (sampled / aggregated).

### ECC / error counters

GPU: ecc.errors.corrected.volatile.*.\<n\>.*[count] — Corrected ECC errors seen in volatile window for subcomponent (cumulative in window).

GPU: ecc.errors.corrected.aggregate.*.\<n\>.*[count] — Aggregate (long-running) corrected ECC errors for subcomponent (cumulative).

GPU: ecc.errors.uncorrected.volatile.*.\<n\>.*[count] — Uncorrected ECC errors in the recent window — critical to monitor (cumulative in window).

GPU: ecc.errors.uncorrected.aggregate.*.\<n\>.*[count] — Aggregate uncorrected errors since stats origin (cumulative).

### Retired pages / reliability

GPU: retired_pages.sbe.\<n\>.*[count] — Pages retired due to single-bit errors (cumulative).

GPU: retired_pages.dbe.\<n\>.*[count] — Pages retired due to double-bit errors (cumulative; more severe).

GPU: retired_pages.pending.\<n\>.*[count] — Pages pending retirement (cumulative / state count).

### Temperature / Power / Clocks / Protected memory

GPU: temperature.gpu.\<n\>.*[C] — GPU die temperature (°C). (sampled)

GPU: temperature.memory.\<n\>.*[C] — GPU memory temperature (°C). (sampled)

GPU: power.draw.\<n\>.*[W] — GPU power draw (watts) for the sampling period (sampled).

GPU: power.draw.average.\<n\>.*[W] — Average power draw (watts). (sampled/aggregate)

GPU: power.draw.instant.\<n\>.*[W] — Instantaneous power draw reading (watts). (instant)

GPU: clocks.current.graphics.\<n\>.*[MHz] — Current graphics clock frequency (MHz). (sampled)

GPU: clocks.current.sm.\<n\>.*[MHz] — Current SM (compute) clock frequency (MHz). (sampled)

GPU: clocks.current.memory.\<n\>.*[MHz] — Current memory clock (MHz). (sampled)

GPU: clocks.current.video.\<n\>.*[MHz] — Current video engine clock (MHz). (sampled)

GPU: protected_memory.total.\<n\>.*[MiB] — Total protected memory (MiB). (instant)

GPU: protected_memory.used.\<n\>.*[MiB] — Protected memory used (MiB). (instant)

GPU: protected_memory.free.\<n\>.*[MiB] — Protected memory free (MiB). (instant)

## Network (NetworkInterStat_diff: — per-interface; \<Interface\> placeholder)

NetworkInterStat_diff:* fields are typically difference/delta values for the sample window (i.e., bytes/packets seen during the interval). Convert to rates by dividing by sampling duration if needed.

### Per-interface fields

NetworkInterStat_diff:rx_bytes.\<Interface\>.*[Bytes] — Bytes received on \<Interface\> during sample window (delta).

NetworkInterStat_diff:rx_packets.\<Interface\>.*[count] — Packets received during window (delta).

NetworkInterStat_diff:rx_errs.\<Interface\>.*[count] — Receive errors in window (delta).

NetworkInterStat_diff:rx_drop.\<Interface\>.*[count] — Received packets dropped in window (delta).

NetworkInterStat_diff:rx_fifo.\<Interface\>.*[count] — RX FIFO errors in window (delta).

NetworkInterStat_diff:rx_frame.\<Interface\>.*[count] — RX frame alignment errors in window (delta).

NetworkInterStat_diff:rx_compressed.\<Interface\>.*[count] — Compressed Rx packets in window (delta).

NetworkInterStat_diff:rx_multicast.\<Interface\>.*[count] — Multicast Rx packets in window (delta).

NetworkInterStat_diff:tx_bytes.\<Interface\>.*[Bytes] — Bytes transmitted on \<Interface\> during window (delta).

NetworkInterStat_diff:tx_packets.\<Interface\>.*[count] — Packets transmitted during window (delta).

NetworkInterStat_diff:tx_errs.\<Interface\>.*[count] — Transmit errors in window (delta).

NetworkInterStat_diff:tx_drop.\<Interface\>.*[count] — Transmit drops in window (delta).

NetworkInterStat_diff:tx_fifo.\<Interface\>.*[count] — TX FIFO errors in window (delta).

NetworkInterStat_diff:tx_colls.\<Interface\>.*[count] — TX collisions in window (delta).

NetworkInterStat_diff:tx_carrier.\<Interface\>.*[count] — Carrier errors on TX in window (delta).

NetworkInterStat_diff:tx_compressed.\<Interface\>.*[count] — Compressed Tx packets in window (delta).

### Network utilization aggregated

NetworkInterStat_diff:NetworkUtilization.\<Interface\>.*[%] — Interface bandwidth utilization percent for the sample window (sampled / derived).

### Protocol & connection counts

NetworkProtocolCounts:TCP.*[count] — Number of TCP connections (instant or sampled depending on collector).

NetworkProtocolCounts:UDP.*[count] — Number of UDP endpoints/sockets (instant).

NetworkProtocolCounts:UNIX.*[count] — Number of UNIX domain sockets (instant).

NetworkProtocolCounts:RAW.*[count] — Raw sockets count (instant).

NetworkProtocolCounts:DCCP.*[count] — DCCP sockets count (instant, if applicable).

NetworkProtocolCounts:SCTP.*[count] — SCTP connections count (instant, if applicable).

NetworkProtocolCounts:local_to.*[count] — Local-to connection count (collector-defined semantics).

NetworkProtocolCounts:to_local.*[count] — To-local connection count (collector-defined semantics).

NetworkProtocolCounts:total_connections.*[count] — Total tracked connections (instant / sampled).

### TCP retransmission metrics (delta)

NetworkTCPRetrans_diff:TcpRetransSegs.*[count] — TCP retransmitted segments in the sample window (delta).

NetworkTCPRetrans_diff:TcpExtTCPSynRetrans.*[count] — TCP SYN retransmissions in window (delta).

NetworkTCPRetrans_diff:TcpExtTCPLostRetransmit.*[count] — Retransmissions that were subsequently lost (delta).