In simple terms, Computational Fluid Dynamics (CFD) is a fluid mechanics discipline that uses mostly Navier-Stokes equations to analyze fluid flow problems. It looks something like this:
I remember attempting to solve these equations with a paper and pencil (and a TI-85) in college- no, thank you. Now that we can use computers to perform these daunting calculations, CFD’s are more pleasant. They are truly a great tool. If you are a gamer, then you see these equations get solved a lot as they are used in video games to model wind, fire and a wide variety of natural phenomena. In engineering, CFD’s are widely used in the design process of structures such as an airplane and its wings.
CFD’s allow you to analyze and visualize the results of a given problem/scenario – but the results depend heavily on the inputs/variables. How CFD’s work? – in general the simulation solver breaks whatever you are trying to model into tiny pieces (iterations) and solves the NS equations for each. After the model is solved, the CFD tool gives you quantitative and visual results data of velocities, temperatures, pressure and other variables specific to the application. There are a few CFD products that are specific to data centers; these include 6Sigma, Tileflow, Flovent, Coolsim, among others. I have used Tileflow and 6Sigma and consider both great tools. Currently I am using 6Sigma at Compass.
For me, its all about having a plan. CFD models in the data center industry are a must in the design phase and could be a should (nice to have) in operations.
A must in design
Consider this, say you are designing a data center in a city like Denver, where air density and altitude are major environmental factors. Try answering these critical questions if you do not use a CFD tool.
[list type=”check”]
- How would you spec out a mechanical unit size without over-engineering it and over-spending – specific to a data center application?
- How would you know (or at least have an educated idea) how your, or your customers, racks perform in a N scenario or in maintenance mode?
- What densities can your data center handle before the need of containment?
- How could you predict hot spots?
- How could you predict negative pressures?
- and so on…
[/list]
I rest my case.
A should (or nice to have) in operations
If you already have a data center in operation, then it would be great to leverage CFDs. The theory is, CFD models do not have to die after the design phase. In operations, where you are doing tons of move/adds/changes constantly, you will benefit tremendously from being able to manage your data center environment more efficiently. Where should I place this 20kw rack? Hmm, run a CFD and you will have a better idea. I know that Future Facilites, the makers of 6Sigma, sell a product specifically for operational management, 6SigmaFM. I haven’t used it, but the concept of using the same CFD that you used in the design phase to plan, test and/or predict your data center’s growth and integrates with your DCIM tool, is very attractive. Beats having to use a thermal imager every time you want to do a move/add/change.
What do you think?
CFD has recently become more available due to stronger computers and reduced software costs, this in turn has made CFD a potential added services for M/E/P firms. Many of the questions you ask can be answered without CFD software, but the effort is significent, and relies on detailed measurements, calculations and some deep experience – but it can be done. As an architect, I became aware fo CFD when more clients requested design validation and we made the decision to obtain the modeling software. I use the tool in many phases of the project from initial conceptual design to alternative layouts and for client support. Most improtant of all is not the software but the ability to use the tool to benefit the overall project. Knowing the implications that are represented and how to move from a less successful solution to a more successful solutions is done be the designer, not the software – its just a tool.
David, thank you for your comment. I agree, we can definitely solve those questions manually in the design process by diving into a soup of numbers and formulas with the right people. Like you said; CFD, at the end of the day, is just a tool. That said, CFD’s are my new TI-85.
As Jose pointed out, the benefit of CFD is that it allows you to simulate any number of scenarios which allows you to make design choices knowing how that will affect the datacenter.
This same idea can apply to data center operations. No datacenter is going to conform to initial design assumptions, and they shouldn’t. However, these deviations to the original design will result in lost capacity. This is where having a predictive model (for space, power, and cooling) has it’s greatest benefits. In this role, CFD is used to fully utilize the cooling that the datacenter is designed for.
We have a webcast coming up talking about exactly how that works:
http://www.futurefacilities.com/media/requestedarticle.php?id=118
Tom, good points. I have readers very interested in the 6SigmaFM concept. I am working with Christian P. @ FF and will set up a lunch-n-learn session in the near future to discuss it.
Have used Coolsim for a couple of years now. It can give an idea of how additions of equipment will effect the ecology of the room. Had direction 2 years ago to install an additional CRAC unit in the room. Ran a model and determined that we had sufficient cooling capacity and did not need the additional CRAC. The CFD model gives you the credibility for your decision. Also helps to determine future capability to support new efforts in the Data Center. Wish I had a CFD when we originally built the Data Center, certainly could have done some things better.
Frank, thanks for your comment. I have heard similar experiences from other end-users. What type of issues where you experiencing initially that made you look at adding the additional CRACs? If the theory was escaping cooling capacity, how did you mitigated it?
Initially it was an expectation of growth and a manager who felt we would need additional cooling capacity.
We’ve done the blanking panels, hot/cold aisles, blocking cable holes in the floors, etc…
I’m looking at whether to consider hot or cold containment.
Did a model on hot aisle containment (my room’s layout seems to lend itself to that).
In the process of building a model for cold aisle containment next to see what results I get.
Frank, it seems that you are doing all the right things. I suggest you also take a look at partial containment. It may be an inexpensive solution that would provide you with immediate results. I always recommend cold aisle containment (when layout allows it) because you have more control of you operation cost equation. If you do it right you can increase the supply temp of your mechanical system and see the dollars shave off your electricity bill. You can definitely accomplish that with hot aisle containment, I just prefer cold.
I’ve used CoolSim for a few years now myself. As an undergrad I worked with a couple other students to analyze the data center on campus in 2008-2009. We left the facilities department with a working baseline model and several alternative models. They ended up implementing our suggestion for cold-aisle containment and were very happy with the results. Recently as a consultant, I used CoolSim to validate a data center cooling strategy designed by one of my colleagues. We were able to identify air mixing issues in the common return air plenum that cannot be fully realized without CFD. We designed in a channel in the plenum to force 100% mixing of the return air. This created uniform loading on each of the CRAC units and eliminated hot spots within the room. As long as we understand CFD is a guideline/tool (not exact), I think it can be useful in a variety of ways.
Dana, thanks for sharing. Another great example of the insight/visibility a CFD can provide. Nicely done.
Hi All
I am an industrial architect with years of experience in data centre airflow dynamics, but have never used CFD due to the huge annual software license cost. Knowing how clients try to screw your fees down, I wonder how they will react to considerable additional charges for CFD. I would love to get into modelling and have a discussion going in LinkedIn to hear from others if enough clients are prepared to pay for this service to allow me to make the investment. Let me know your thought.
Jose, I’m a total advocate of CFD and use the 6SigmaDC software on an almost daily basis as a consultant at Dell. I’ve been working with CFD for almost 15 years now and every time I see the Navier-Stokes equations a shiver runs down my spine. I’m sure I’m not alone as many (like me) will have had a brush-in with these bad boys during their Engineering studies.
Thankfully you really don’t need to know this to use the better tools and to get good results. In the same way that you don’t need to know how the engine works in order to drive your car. For this reason I’d recommend opening your blog more gently to avoid frightening off anyone without a math phd who was considering embracing a CFD tool for their datacenter for the first time.
Stu
P.S. Love the blog. Shout out if you need some Datacenter CFD images for the site.