All webinars

Simulating Energy Utilization in the Data Center

Why PUE is a weak efficiency metric, how the 90.4 mechanical load component works, and how CFD feeds an annual energy model.

This webinar shows how to simulate the annual energy use of a data center by coupling a computational fluid dynamics (CFD) airflow model to a cooling plant energy model. It is aimed at data center designers, operators and energy practitioners who want a better measure than Power Usage Effectiveness (PUE) and a way to test changes before building them.

What the webinar covers

The two presenters start with the criticisms of PUE: it should be an annual figure, it is hard to measure because IT load varies, it ignores climate, and it can be gamed. They argue that cooling power is the dominant term above the IT load. They then describe ASHRAE Standard 90.4, whose mechanical load component is a system-level ratio of cooling power to IT power. It is annualized by climate zone and evaluated at four load points, so it shows how the cooling system behaves at low IT load.

The core idea is to give each fan, pump, chiller and cooling tower an empirical efficiency curve that depends on part load and outdoor conditions, and to calculate the plant energy for every hour of a year from weather data. CFD output from the room feeds that calculation. The presenters say this makes it possible to compare climates, cooling system types, control strategies and failure cases.

The demonstration uses two chilled water rooms served by air-cooled chillers with an economizer loop. Both start with too much air and too-cold rack inlets, and the CFD plots show recirculation and one room with racks too close to a cooling unit. Containment, matching supply air to rack demand and a ten degree Fahrenheit higher supply temperature improve both rooms. The new energy report is shown, with monthly and hourly plots. A question-and-answer section covers metering, pumping, cooling towers and custom modeling services.

Key points

  • PUE moves little when only cooling improves, because IT load dominates the ratio. A cooling-power ratio such as the mechanical load component responds more clearly.
  • A design sized for peak load often runs far below it, so efficiency at 25, 50 and 75 percent load matters as well as at 100 percent.
  • Component efficiency depends on part load ratio and weather, so an annual hour-by-hour calculation is needed. An air-cooled chiller depends mainly on dry bulb temperature.
  • In the example, airflow management alone (containment, matched airflow, warmer supply air) cut modeled cooling power by about a quarter and annual cost by roughly ten thousand dollars. PUE changed far less.
  • Mechanical plant changes, such as variable frequency drives on pumps, were described as a larger opportunity, but they were not modeled in this example.
  • For monitoring, the presenters suggest rack-level inlet and outlet temperatures, sampled on the worst racks, and comparing measured data against a calibrated model.