Skip to content

Power consumption (Linux servers)

Monitoring the power consumption of a Linux machine is critical for optimizing resource use, maintaining energy efficiency, and supporting sustainability efforts. Below, we explore various methods and tools to measure real-time power consumption from the server’s outlet, CPU, or other hardware components.

Overview

This guide outlines methods to measure the real-time power usage of a Linux system. We’ll explore techniques using built-in hardware sensors, software tools, and external power meters to gather precise measurements of power consumption. These methods apply to servers, desktops, and laptops configured with Linux.

Methods for Measuring Real-Time Power Usage

Using the powercap Interface and RAPL

Intel’s Running Average Power Limit (RAPL) framework provides detailed power consumption metrics for CPUs, DRAM, and GPUs. This is particularly useful for modern servers equipped with Intel processors.

  • Checking Instantaneous Power Usage:

    bash sudo cat /sys/class/powercap/intel-rapl/intel-rapl:0/energy_uj

    • The value represents energy consumption in microjoules.

    • To calculate power (in watts), take two readings over a time interval and compute:

      bash time=1 T0=$(cat /sys/class/powercap/intel-rapl/intel-rapl:0/energy_uj) sleep $time T1=$(cat /sys/class/powercap/intel-rapl/intel-rapl:0/energy_uj) echo "scale=2; ($T1 - $T0) / ($time * 1e6)" | bc

  • Tool Integration: The turbostat utility, included in the linux-tools-common package, also provides power consumption information.

    bash sudo turbostat --Summary

Monitoring with lm-sensors

lm-sensors provides access to voltage, temperature, and power sensors embedded in the server hardware. It is particularly effective for servers with supported ACPI or motherboard sensors.

  • Installation:

    bash sudo apt install lm-sensors sudo sensors-detect

  • Usage:

    bash sensors

    • Look for power metrics like power1.

GPU Power Monitoring

For servers equipped with discrete GPUs, power consumption can be measured using vendor-specific tools:

  • NVIDIA GPUs: Use the nvidia-smi tool:

    bash nvidia-smi --query-gpu=power.draw --format=csv

  • AMD GPUs: Use rocm-smi:

    bash rocm-smi --showpower

Using Tools Like powertop and powerstat

  • Powertop: Provides real-time power estimates and a breakdown by process:

    bash sudo powertop

    • Offers detailed insights into power-intensive processes and suggestions for power optimization.
    • Powerstat: Tracks and logs power consumption:

    bash sudo apt install powerstat powerstat -d 0 1 60

    • Customizable for intervals and sampling frequency, providing averaged and peak power data.

Advanced Power Profiling with External Tools

For precise and reliable power usage metrics, external power meters such as Kill A Watt or intelligent PDUs are ideal. These devices monitor real-time power draw from the power outlet, accounting for total system consumption, including inefficiencies in the power supply unit (PSU).

  • Setup: Plug the server into the power meter, which in turn connects to the power outlet.
  • Benefits:
    • Provides highly accurate measurements.
    • Includes variance from dynamic workloads and PSU efficiency.
  • Advanced Meters: Some models can transmit real-time data to networked devices for logging and visualization.

Practical Considerations

Factors Impacting Measurements

  • Dynamic Workloads: CPU/GPU load fluctuations can impact readings.
  • PSU Efficiency: Power usage from the outlet is higher than internal component consumption due to PSU inefficiencies.

Limitations of Software-Based Methods

Software tools rely on hardware sensors and may not account for total power usage. They are best used for profiling specific components, like CPUs or GPUs, rather than total system power.

External Meters for Accurate Server Monitoring

In scenarios requiring precise total consumption data (e.g., for billing or capacity planning), external meters are essential.

Conclusion

Monitoring power consumption in Linux servers can be accomplished using various tools and methods, depending on the desired level of accuracy and the specific use case. While software-based tools like powercap, lm-sensors, and powertop provide quick and accessible insights, external hardware solutions are necessary for precise measurements. By leveraging these techniques, administrators can better manage energy costs, improve efficiency, and contribute to sustainable operations.

Page last modified: 2024-11-25 08:54:17