Linux rapl metricset

The RAPL metricset reports power usage in Watts as reported by Intel’s RAPL (Running Average Power Limit) API. In practice, RAPL is used for setting power usage limits on a CPU, but it can also be used for monitoring power usage.

RAPL is available on most Intel CPUs in some form since Sandy Bridge, however, this implementation requires a linux distro that exposes intel MSRs (model-specific registers) via /dev/cpu/[CPU]/msr. You can check to see if your linux distro and CPU support RAPL by running

sudo rdmsr 0x606

If that command returns a hex value such as a1003, RAPL is supported.

RAPL divides a CPU into four domains, each one corresponding to a component within the hardware itself: Package, DRAM, PP0 (usually the collective power usage of the cores), and PP1 (usually the integrated GPU). Not all packages are available on all CPUs, and the RAPL metricset will report all RAPL domains that it finds.

This metricset also supports msr-safe, which allows a user to access an MSR device without root. The rapl.use_msr_safe flag in the linux module config will enable usage of the /dev/cpu/[CPU]/msr_safe device.

For existing msr-safe installs, the following allowlist will open all RAPL MSRs for reading:

0x00000606 0x0000000000000000 # "SMSR_RAPL_POWER_UNIT"

0x00000610 0x0000000000000000 # "SMSR_PACKAGE_POWER_LIMIT"
0x00000611 0x0000000000000000 # "SMSR_PACKAGE_ENERGY_STATUS"
0x00000613 0x0000000000000000 # "SMSR_PACKAGE_PERF_STATUS"
0x00000614 0x0000000000000000 # "SMSR_PACKAGE_POWER_INFO"

0x00000618 0x0000000000000000 # "SMSR_DRAM_POWER_LIMIT"
0x00000619 0x0000000000000000 # "SMSR_DRAM_ENERGY_STATUS"
0x0000061b 0x0000000000000000 # "SMSR_DRAM_PERF_STATUS"
0x0000061c 0x0000000000000000 # "SMSR_DRAM_POWER_INFO"

0x00000638 0x0000000000000000 # "SMSR_PP0_POWER_LIMIT"
0x00000639 0x0000000000000000 # "SMSR_PP0_ENERGY_STATUS"
0x0000063a 0x0000000000000000 # "SMSR_PP0_POLICY"
0x0000063b 0x0000000000000000 # "SMSR_PP0_PERF_STATUS"

0x00000640 0x0000000000000000 # "SMSR_PP1_POWER_LIMIT"
0x00000641 0x0000000000000000 # "SMSR_PP1_ENERGY_STATUS"
0x00000642 0x0000000000000000 # "SMSR_PP1_POLICY"

{
    "@timestamp": "2017-10-12T08:05:34.853Z",
    "event": {
        "dataset": "linux.rapl",
        "duration": 115000,
        "module": "linux"
    },
    "linux": {
        "rapl": {
            "core": 0,
            "dram": {
                "joules": 38106.59,
                "watts": 5.912
            },
            "package": {
                "joules": 25764.9984,
                "watts": 11.5784
            },
            "pp0": {
                "joules": 27353.5644,
                "watts": 2.7159
            }
        }
    },
    "metricset": {
        "name": "rapl",
        "period": 10000
    },
    "service": {
        "type": "linux"
    }
}