All webinars

Using CFD Modeling to Analyze High Density Data Centers

What a CFD study of a hot-aisle-contained, non-raised-floor data hall shows about hot aisle width, ceiling height and containment leakage.

This webinar uses a generic, non-raised-floor data hall with hot aisle containment to show how computational fluid dynamics (CFD) reveals problems that simple airflow arithmetic misses. It is aimed at designers and operators of high-density halls that are flooded with cold air from one side.

What the webinar covers

The presenters begin with the design: air supplied from one wall (for example from rooftop or free-cooling units), racks in contained hot aisles, and exhaust through the ceiling. The model covers only airflow in the white space and holds the supply temperature fixed, which lets the supply temperature be raised to test a worst-case day. It is a steady-state model, and the presenters explain that real control systems vary over time.

The pre-simulation report compares the airflow the cooling units supply with what the racks demand. The demand comes from the convective heat transfer equation, and the presenters discuss how much the temperature rise across IT equipment varies between models and over time.

The surprise in the first results is that pressure is higher in the hot aisles than in the cold aisles, the opposite of what a sealed, well-designed containment should show. The presenters test three explanations one at a time: colder supply air, wider hot aisles, and a taller return plenum. Supply temperature has little effect. Wider aisles reduce the pressure difference. A taller plenum reduces it the most.

They then add containment leakage, first as a percentage and then as small open areas in the racks built with a spreadsheet-style rack builder, and clone the rack row across the hall. The webinar ends with a run of cooling-unit-off variations and a discussion of whether rate-of-rise can be modeled.

Key points

  • Give the hot side enough volume. Air leaving the racks has to turn and rise, and that momentum raises pressure in a narrow hot aisle.
  • Geometry on the hot side mattered more than supply temperature in this example, and ceiling height mattered most.
  • A leakage figure such as “5 percent” is ambiguous: of airflow, of containment area or of rack area. The presenters prefer to place the leakage where it occurs, such as around and through the racks.
  • Start with a coarse model and add detail only where it is needed, because detail costs solve time.
  • Run the cooling-unit-off cases for every unit, since position can make some units much more important than others.
  • Rate-of-rise (transient) analysis iterates every time step to convergence, much like a small steady-state case, so it costs far more computing time than a single steady-state run.