University

Server Power Consumption and PUE

Server Power Consumption and PUE

Definition

Server power consumption includes a substantial baseline even at low CPU utilization. Power Usage Effectiveness (PUE) relates the whole datacenter’s energy use to the energy used by its IT equipment.

What the lecture shows

  • Slide 8: a large sleep-to-idle power increase, then additional consumption as CPU utilization rises.
  • Slide 9: different machines draw different power at comparable PassMark scores. Its horizontal axis is performance score, not utilization.
  • Slide 10: a reported typical utilization of 5-15% motivates consolidation, subject to contention and QoE limits.
  • Slides 11-12: illustrative hardware power, rack density, cooling, and facility overhead figures, including examples labeled 2008.

The point is that idle hardware still costs energy. Moving workloads can help if enough physical machines can then be powered down or put into a lower-power state.

Study model: baseline power

For an illustrative server, use:

P(u) = P_idle + (P_peak - P_idle) × u

Here u is utilization from 0 to 1. This is a simplified explanation, not a universal measured power curve.

Suppose idle power is 100 W and peak power is 200 W. Three servers at 10% utilization use 3 × 110 = 330 W. One server at 30% uses 130 W. If the other two are fully powered off, this model saves 200 W. This assumes compatible workloads, sufficient capacity, and ignores additional virtualization and facility effects.

Study clarification: PUE

PUE = Total facility energy / IT equipment energy

Both values must use the same measurement boundary and period. The equivalent power ratio can be used for consistently measured power over the same conditions.

For 180 kW facility power and 100 kW IT power:

PUE = 180 / 100 = 1.8
Non-IT overhead = 180 - 100 = 80 kW

The lecture’s multipliers of approximately 3, 1.8, and 1.2 illustrate how facility overhead changes total consumption. They are lecture examples, not current industry benchmarks.

Common mistakes

  • Assuming 10% CPU utilization means 10% of peak power.
  • Reporting watts as energy; watts measure power, while energy over time can be expressed in Wh or kWh.
  • Treating a lower PUE as proof of lower total energy consumption or more useful computation. PUE describes facility overhead relative to IT energy.
  • Comparing performance or power alone and ignoring the other axis.

Source

Fulvio Risso, Introduction to Computing Virtualization, slides 8-12. Lecture context: Lecture 02 - Introduction to Computing Virtualization. The formulas, worked calculations, and measurement cautions are additional study explanations; the slides introduce PUE without writing the ratio explicitly.

Graph View

Backlinks