Issue link: https://cp.revolio.com/i/148754
Often, owners ask for certain types of analyses not realizing that the request will not answer their questions. More importantly, there may be important information contained within each of these applications that are not asked for nor fully utilized by the owner. CFD The most common application for CFD modeling is to obtain temperature profiles throughout the data hall under design conditions. One aspect of modeling that is not often thought about is that CFD is of little use when applied to contained environments. True containment, i.e. contained environments in which the containment is rigid and there is little or no leakage, leaves little-to-no possibility of mixing air streams in either the cold aisle or the hot aisle. The supply air temperature WILL be the same as the cold aisle temperature, and the return air temperature WILL be representative of the AVERAGE of the server discharge temperatures in the hot aisle. If an owner is looking for the inlet temperature profile and the space is contained, little or no new information will be obtained by performing the CFD modeling. Furthermore, the method of air delivery, whether from above or below, whether from open ducts or floor tiles, is irrelevant (from a CFD standpoint) provided that the system air flow matches the server air flow. This matching of air flow is a necessary condition for containment; if the air flows don't match, there will be either recirculation air flow, bypass air flow, or both... meaning that the space is not contained to begin with. On the other hand, if pressure profiles are desired for the purpose of identifying effective placement of pressure sensors, useful information can be obtained in the analysis and has good application in determining the best control strategy for the system fans to track with the server fans. For the most part, all further discussion of CFD modeling will refer to uncontained environments. CFD modeling is best applied to test the effectiveness of alternate system configurations. There is little point in running a single model without comparing it to an alternate solution since the probability is almost nil that it will match the real data hall to the smallest detail once fully built and developed. Just as in other types 18 | THE DATA CENTER JOURNAL of computer simulations or modeling schemes, the point of the model is to find the better performer under specific conditions of growth and/or load density; the model should not be assumed to be able to closely predict reality. The modeler may want to model alternate systems, such as overhead air distribution versus underfloor air distribution (the macro level); trying to include details of where the holes are placed within a perforated tile is wasteful (the micro level). In an as-built situation, it is usually sufficient to model "well enough" in order to compare it to a possible improved condition. Although the macro-micro split is relative to the problem being considered, it is important to draw a line between them in order to model the macro conditions and avoid the micro details. Frequently, the improved conditions can be identified and implemented using best practice procedures, which can be found from sources such as the numerous publications by ASHRAE Technical Committee 9.9 Mission Critical Facilities, Data Centers, Technology Spaces and Electronic Equipment. Too often owners and operators don't trust the best practice techniques because they may seem "rule-of-thumb." What is missed in their approach is that their own proposed solutions originate from "seat-of-thepants" engineering, which is not only less dependable than the best practices, but it's often contradicted by CFD modeling. The human mind is not capable of predicting how air will flow when subjected to numerous obstacles and impediments to flow in a critical environment. Therefore, the best approach is to apply the best practices first before doing any modeling. If it's an existing condition with air flow problems, the owner is best served by trying to implement the best practices first; only then, if the problems persist, apply the CFD modeling techniques to take the problem solving strategy to the next step (i.e. proof of concept). If the CFD modeling were performed first, it would model a system in the as-is condition, and the next logical step would be to model it with the best practices applied. That's wasteful of time and money. There are a few aspects of CFD modeling that are misunderstood. Understanding these simple concepts can lead to better decision making processes for the owner. One aspect that needs to be explained is that a CFD model showing temperature profiles and/or particle traces, though it makes for a pretty picture, does not represent an instant in time. Turbulence and eddy currents occur in fleeting moments in real time and cannot be captured in CFD (unless transient analyses are made, which is orders of magnitude more complex than a static model). If one could take a long time exposure photo of the quantities being looked at in a CFD model, it would eventually come to look like the static picture created by the CFD model. Therefore, one should not be alarmed if a CFD model does not match exactly the measurements taken in the field. Another aspect that is misunderstood is that when alternate solutions are tested in a model, the result of the analysis will often be alternate temperature profiles across the entire face of the server inlets. What was perceived to be a temperature extreme in Option 1 may be solved in Option 2, only to be replaced by another temperature extreme at a different server or cabinet. Which is the better solution? Again, the human mind is not often capable of processing complex temperature/color distribution profiles in order to make a decisive call about which is the better option. In these circumstances, CFD allows virtual temperature sensors to be placed at numerous locations within the modeled space, and these virtual temperature readings can be used to generate a number representative of a metric that describes how much mixing occurs in the space or how many servers fall within a specific temperature range specification. Using metrics in this manner can remove some of the subjective aspects of interpreting the results of the CFD modeling. (A discussion of metrics such as SHI/RHI (Supply Heat Index/Return Heat Index) or RCI (Rack Cooling Index) are beyond the scope of this article.) Don't rely on CFD modeling for all solutions. Apply best practices first. Apply CFD model to systems that already have the best practices applied. CAD/BIM Computer Aided Design (CAD) has been the standard for the AEC industry for over 20 years. CAD provided a means of expediting the design and drafting process by providing digital design tools to quickly layout designs and make revisions. Multiple systems or disciplines are www.datacenterjournal.com