All webinars

Creative Cooling Strategies for Existing Data Centers

A CFD case study of an older 6,000 square foot raised floor data hall: can it grow past a megawatt with a shallow, cluttered underfloor?

This webinar walks through a computational fluid dynamics (CFD) study of an older, irregularly shaped data hall with a shallow raised floor. It is aimed at operators and engineers who have to add load to an existing room without rebuilding it. Two presenters discuss the study: the president of Applied Math Modeling, which develops CoolSim, and a data center industry veteran.

What the webinar covers

The base model is a roughly 6,000 square foot hall with a 15 inch raised floor, no return plenum, and pipes and cable trays under the floor. It is cooled by computer room air conditioner (CRAC) units around the perimeter and relies on natural convection for the return air. The presenters review the model’s reports, including the rack inlet plots drawn against the ASHRAE recommended range.

They then look at the underfloor. Air leaves the CRAC units at high velocity, so it does not turn up into the nearest tiles. Removing the pipes and trays made the tile airflow worse in this model, and swapping in higher-open-area tiles in a few places changed flow at tiles several positions away.

The customer scenario is growth to a little over a megawatt with N+1 redundancy. The existing units cannot supply enough air, so the model replaces them with newer units that have direct-drive electronically commutated (EC) fans. The presenters check the result with Rack Cooling Index (RCI), Return Temperature Index (RTI), underfloor pressure and the fan curves. They then turn off one cooling unit at a time to test failures.

The same hall is rerun with hot aisle containment, and then with the cooling units raised on pedestals to make a non-raised-floor layout. The closing questions cover underfloor air velocity, finned tiles, blanking panels and combining hot and cold aisle containment.

Key points

  • A shallow or cluttered underfloor does not rule out more load. In this model the obstructions slowed the air near the CRAC units and made tile airflow more even.
  • Tile changes are trial and error. One higher-open-area tile can change the airflow at tiles several positions away, which is why the presenters model the change first.
  • Check airflow as well as tons of cooling. The presenters suggest about 156 cubic feet per minute (CFM) per kilowatt at a 20 degree F rise across the servers, and a little over 100 at a 30 degree F rise.
  • Test failures one unit at a time. With all units running the upgraded hall looked acceptable, but with a unit off some racks ran too hot while others ran too cold.
  • Hot aisle containment let the model raise the supply temperature and kept the room cooler, but tight containment gives servers less room to recover when a cooling unit fails. The presenters do not recommend hot and cold containment in the same design.
  • Raising the units on pedestals worked in normal operation but failed some single-unit shutdowns, because the air jammed against the back of the containment. The presenters say this was not obvious before running the models.
  • The presenters describe CFD as a way to eliminate bad ideas and support engineering decisions, not as a substitute for them, and as a tool operators can keep using after the design stage.