Fan Efficiency and Economies in the Data Center: The Rest of the Story | Chatsworth Products
 

Cross-Connect Blog

Fan Efficiency and Economies in the Data Center: The Rest of the Story

Fan Efficiency and Economies in the Data Center: The Rest of the Story

 Permanent link

Mark Fontecchio has written the first intelligent article on fan efficiency and economies in the data center that I have read. His article, “Data Center Air-Conditioning Fans Blow Cost Savings Your Way,” can be found under Data Center News (January 21, 2009) on the SearchDataCenter.com website. I encourage you to read this article, but then you need to come back here for the “rest of the story” – even though Mark has really captured the core information on fans, some of his sources were not completely forthcoming, and I am more than happy to address those calculated omissions here.

So … you have read Mark’s article and now have a good understanding of plug fans and variable frequency drive (VFD)-equipped fan motors. Now, for the rest of the story. While it is true that air handlers with VFD will provide a great economic benefit during the low utilization stages of a data center ramp-up, that is not where the benefits end. These air handlers provide an ongoing benefit through energy efficiency gains from simultaneously running fail-over redundancy. With non-VFD air handlers, your redundancy is achieved either by over-provisioning all the time or by having some number of CRAC units not running and in ready-reserve. With VFD CRAC units, you would run your extra units at reduced capacity and reap the benefits.

For example, if your space required the sensible cooling capacity of five 30 ton CRAC units, your N+1 availability would give you six units; your N+2 would give you seven units and your 2N would give you ten units. In N+1, you would be running each unit at 83% capacity and, with the cube effect described in Mark’s article, each CRAC unit would be consuming 57% of full capacity energy and with six units running instead of five, that still comes out to a 32% energy savings. In N+2, each CRAC unit is pushing air at 71% capacity while therefore consuming only 36% of the energy (.713) and with seven CRAC units running, that results in a total 50% energy savings. Finally, for 2N uptime availability, you would have ten CRAC units pushing air at 50% capacity while consuming 12.5% of the energy each, for a total 75% energy reduction over those ten CRAC units versus five units at 100% capacity.

Obviously, the economic benefit of VFD does not stop delivering at full deployment – that’s the gift that just keeps on giving. The proposition for long payback periods for plug fans also needs a little voice of reason. First, the incremental cost for plug fans from CRAC and air handler manufacturers for whom those types of fans are standard offerings, is much less than cited in the article. More importantly, in addition to their inherent efficiency superiority, they also provide additional benefits that ought to be considered in any ROI analysis. For example, because their output is high pressure and low velocity, the normal practice of having to locate the closest perforated floor tile some 8-10 feet away from the air handler so the air can slow down enough to build enough pressure to actually force some air through that tile does not have to be followed with plug fans. Since perforated floor tiles and hence also server cabinets can be located much closer to the air handlers, you can get much better space utilization – less empty space in the data center.

Furthermore, because of the velocity of air produced by centrifugal fans, data centers so equipped tend to have extremely wide variations of pressure under the floor, resulting in wide variations in airflow through the perforated tiles throughout the room. These variations are compensated for in two ways: either over-provisioning airflow into the room to assure there is adequate air availability in lower pressure areas or reducing the set point to minimize the effect of re-circulated return air in those low pressure areas. By calculating the additional savings associated with these two improvements in a specific application environment, the ROI for plug fans will be much quicker, if not instantaneous.

Finally, those folks at APC and other folks with close-coupled cooling solutions are just going to have to get over it and face that sooner or later they are going to get found out. Granted, when the benchmark is a train-wreck of a legacy data center, those close coupled solutions do compare well. However, they just don’t stand up to a well designed data center with air being moved around by plug fans with VFD.

Here is just a partial list of the more obvious snakes:

  1. Large fans are always more efficient than small fans. (If you’re not a mechanical engineer, just find one down the hall or in your local pub or honky tonk and get your verification)
  2. While VFD on a small close-coupled fan is better than not, it can’t be as efficient as a larger un-coupled fan because you cannot run your N+ redundant capacity in a one-to-one close-coupled model.
  3. The argument that the close-coupled fans don’t have to push the air long distances doesn’t hold up to a real basic scrutiny based on elementary physics. In a well designed data center with maximum segregation between supply air and return air, the air handlers are not pushing air any farther than just into the room. All the flow is created by pressure differentials. The server fans create low pressure that pulls in air, that is replaced somewhere in the room by the air handlers. The air handlers likewise create low pressure that pulls in the return air. As long as these two streams are separated by servers, all the flow is created by pressure differentials and the fans are just compensating for what has been pulled out of the room.

CPI's Passive Cooling Solutions completely segregates supply and return air by using a "chimney cabinet" to direct airflow into a drop ceiling. Find out more at www.chatsworth.com/passivecooling, or call us at 800-834-4969. Of course, you can also leave your comments below.  Ian Seaton, Technology Marketing Manager


Posted by Ian Seaton, Technology Marketing Manager at 10/11/2011 09:32:32 AM




Comments

Ian, thanks for the mention and for your comments. I continue to learn about the importance of air handler fans even after having reported it out. It seems to be overlooked somewhat, but with data center cooling being such a big part of the overall power envelope, and with fan power being a large part of that, it can't be ignored.

On another note, I saw you speak at ASHRAE and meant to introduce myself but didn't have the time. We'll have to meet up in the future. I'm now working on a story about the importance of server racks/cabinets in data center airflow, and will be using info I got from ASHRAE for it.

Take care,
Mark
Posted by: Mark Fontecchio at 1/28/2009 3:47 PM


Interesting to read this article here, as my background is in chemical engineering, where I specialised in fluid mixing.

The reason for my comment is that you are only partially correct when you say that large fans are always more efficient than small fans. reasons are as follows. Note that I am basing this on the flow of an incompressible fluid, air being compressible is slightly more complex: -

1. Equations for pumping capacity Q is Q=Nq.N.D^3 where N is the rotational speed and D is the fan diameter. Nq is the pumping number and is a characteristic of the design of the fan. The other equationa of interest is for power, P. That one is P=Np.rho.N^3.D^5. That means that P/Q=Np/Nq.rho.N^2.D^2. The implication is that efficiency in terms of Power required per unit of flow is a function of the impeller design (Np/Nq), speed and diameter.
2. It is important to understand the difference between efficiency in power per unit volume and efficiency in terms of what you want to do per unit of power. Heat transfer for instance is also effected by turbulence, and eddy size, which is in turn negatively effected by power per unit volume (think of moving air as distribution of power).

Anyway, better get back to work, hope this is helpful in your thinking!
Posted by: funkyg at 1/29/2009 4:24 PM


Plenum resistance seems like a key factor driving efficiency. In many circumstances, this simply devolves to the height of the raised floor. (remoteness of CRAC units and under-floor impediments to free flow of air are also important considerations...)
If we want to significantly reduce the cost of moving the air, we should build computer rooms with very low resistance supply plenums (e.g. 6-foot tall raised floors with no impediments).
Posted by: Robert Enger at 7/7/2009 12:04 AM


Leave a comment
Name (required)
Email (optional, not published)
Your URL (optional)
Comment

 

Note: Conversation is encouraged and expected. However,
moderation of comments is necessary to prevent spam, per-
sonal attacks, profanity, or off-topic commentary. Comments
related to specific product support or customer service issues
will be addressed separately rather than posted here. Please
email techsupport@chatsworth.com for assistance with these
matters.