CFD is a powerful design tool when considering a data center cooling scheme. With CFD you can display contour plots of temperature, speed, and for the purists there is pressure. I, perhaps naively, have always considered pressure to be a means to an end. I can absolutely see the value of pressure in terms of “pressure drop” but for me to look at a contour plot of pressure at a room level and deduce how my design should change makes my head hurt. Don’t hate me. You can also animate the air flow by seeding the flow with neutrally bouyant virtual particles. Very powerful indeed but qualitative in nature. In other words, very useful in painting a picture of the quality of the design to colleagues or management but doesn’t give the designer the full analytical benefit.
To illustrate the value of the various types of CFD outputs for a data center airflow design consider the following scenario. We have a raised floor data center with two downflow room coolers. Each cabinet has a floor tile adjacent to it with the hopes of supplying a dedicated source of cool air. The image below, a temperature contour and velocity vector plot near the inlet to a row of cabinets, illustrates the result of this design. Notice that the air in the upper portion of the cabinets is heated prior to entering the cabinets. Upon closer inspection we also notice that the first two floor tiles are supplying air to the plenum rather than delivering cool air to the caibinets. Not really what we had hoped for.
Another useful output from CFD is the ability to animate the flow which I have shown below. It shows, quite clearly I believe, that we have some flow sneaking over the top of the cabinets rather than heading straight back to the cooler.
A relatively new metric when designing an airflow management solution for a data center has to do with this thing called “Capture Index” (James W. VanGilder, Saurabh K. Shrivastava: Capture Index: An Airflow-Based Rack Cooling Performance Metric, ASHRAE Transactions 2007, DA-07-014, Volume 113, Part 1)
It is a very powerful metric which is worthy of a dedicated blog, and a whole lot more. In my next blog I will illustrate the additional information which can be gleaned from this design by way of the “Capture Index”.