By: Liam Newcombe

When does chasing a low PUE start to cost money instead of saving you money?

There is no shortage of data center products which claim to offer huge reductions in PUE and an associated large ROI. The problem is that, in many cases, chasing greater and greater PUE reductions is the wrong thing to do, both financially and environmentally.

In this two-part post, we will examine one common data center design option that many would expect to not only result in a lower PUE, but to also save a lot of energy and money. Surprisingly, when a detailed analysis is carried out there is no financial or environmental benefit despite reducing the PUE. In fact, for a $500k initial investment the 10 year ROI evaluates to a loss of roughly $400k, the reasons why the apparently obvious choice doesn’t work are discussed below.

The key take-away from this specific case should not be that this type of cooling system doesn’t work, as this is not necessarily a general result. What we can learn however, is that assessment of TCO and ROI for data center investment choices should not be based on weak proxy indicators such as “free cooling hours” as this is likely to include errors which are larger than the estimated savings. Effective analysis of the financial as well as the engineering elements is key to avoid costly mistakes. A specific outcome is that the proliferation of web based and pre-sales guesstimator tools which rely on weak indicators, or worse, do not describe their methodology, are subject to such large errors that operators should think very carefully before basing any decision on their output.

Wouldn’t adiabatic cooling work well in New Mexico?

The example we will use is a new-build 1.2MW IT load data center with direct outside air economisers and two reasonable, but modern IT equipment environmental ranges. The design is a Tier III data center using good quality components which achieves a good balance of first capital cost and Total Cost of Ownership. The question is whether it is worth adding adiabatic (evaporative cooling and humidification) sections to the Air Handling Units (AHUs) which would allow the site to operate on outside air instead of mechanical cooling for many additional hours per year.

• The instinctive response is that adiabatic cooling should show a huge energy efficiency advantage in a dry climate such as New Mexico as long as we contain the, relatively small, water cost
• The problem is that these adiabatic components are relatively expensive to purchase and continue to cost money to maintain.

Our modern design site has two choices of supply air control ranges to keep the IT equipment within the ASHRAE Class A1 range. The table below summarises the control boundaries for those who want to know how the analysis was configured;

Table 1 Environmental control ranges

Our dry design can run entirely “free” cooling only when the outside air is between the humidity limits and below the target supply temperature. The problem for the non-adiabatic design is that the climate in New Mexico is too cold and dry in the winter with too little humidity in the outside air, or hot and dry in the summer, too hot to run free cooling. Under these conditions surely it would make good sense to use adiabatic units to humidify the air when it is too dry and to cool it for free when it is too hot?

It is common to see people try to assess the performance of a free cooling system by comparing how many “free cooling” hours per year it achieves. Intuitively it seems that this would be a good indicator of the performance, more free hours must be better after all.

The table and graph below shows the percentage of the typical year that each version will spend in full free cooling, partial free cooling and full mechanical cooling;

Table 2 Free cooling hours

New Mexico Free, Partial and Full Mechanical Cooling Hours

Looking at the results of the free cooling analysis for our site in the dry New Mexico climate, it would seem that the adiabatic option is a no-brainer and must hugely outperform the standard outside air system.
If we were to try to base an operational cost savings estimate on this free cooling hours analysis we would need estimates of;

• Base power cost $0.0058 / kWh
• IT kW load 1,000kW average
• Cooling overhead fudge factor 0.25 * cooling load

If we use this approach to determine the operational cost savings then our output might look like;
Table 3 Flawed cost savings analysis

Despite appearing to be derived using a reasonable method, the savings estimates above are completely wrong. As we will show in the next couple of sections, due to factors such as the interactions of climate, varying mechanical compressor efficiency with external temperature and part loading, free cooling hours are not a useful indicator of energy or cost. This is described in more detail in our cooling analysis paper developed for the EU Code of Conduct on Data Centers.

PUE comparison

If instead of simply counting the free cooling hours we carry out a more detailed assessment of the achieved performance we can get a better understanding of what is going on. The output below is from a full hourly simulation of both normal (red) and adiabatic (blue) designs operating at both supply air temperatures which takes into account cooling load, varying DX compressor performance etc. (note that for readability the PUE axis starts at 1.0, not 0)

PUE Comparison

From the left hand chart, at 75°F supply, with adiabatic cooling (blue) we can run almost the whole year without starting our mechanical compressors and they do not run at full load for many hours at all resulting in the small spiked in July – September. Without the adiabatic cooling we have more hours of full mechanical cooling, but these are predominantly in the winter when the outside air is too dry to supply to the data hall.

In the right hand charts, at 58°F even with adiabatic cooling the data center relies heavily on mechanical cooling in the summer where the outside air is too hot or too humid. The normal, dry cooling option at 58°F on the other hand, has few days where it does not use the mechanical cooling, quite a few days of mechanical cooling during the day in the hot summer plus the days of full mechanical cooling in the cold-dry winter conditions.

Water consumption

Given the selection of New Mexico, a region with little water and therefore relatively high water prices, we need to check the water consumption in addition to the energy profile to determine the environmental and financial performance of our design options. The adiabatic humidifier sections use water in two operating modes for our site;

• Humidification of outside air that is too dry to meet the minimum IT humidity target
• Cooling of outside air that is too hot to meet the maximum IT supply temperature target

The water consumption is calculated by determining the water requirement to meet the change in air moisture conditions. A 50% overhead is then added to this to account for water lost from the system due to flushing and other processes, this is quite conservative but depends greatly upon the individual units in use, anything up to 200% overhead is realistic. A water cost of $22 per 1,000 Gallons as an indicative sample price the water consumption to achieve the minimum supply humidity and adiabatic cooling targets is shown below;

Table 4 Annual water consumption of adiabatic humidification and cooling

TCO over 10 years

Given the relatively small water cost there is still some hope for a good TCO outcome on the adiabatic option.

To perform the TCO analysis we will;

• Evaluate both options over 10 years with a 7% (adjusted) discount rate
• Assume flat (real) water and power costs over the period
• Apply a $500,000 first capital cost difference to build with the adiabatic sections in the AHUs based on actual price data
• Apply a $10,000 per annum operational and maintenance cost for the cleaning and treatment of the adiabatic cooling chambers
• Allow the IT power draw to rise over the first four years as 250kW, 500kW, 750kW and then 1,000kW for the remaining years
• $0.058 per kWh average utility energy cost

Putting these numbers into the TCO analysis we can now plot our TCO over 10 years to see what overall saving we are able to recover against the initial capital investment from reduced operational cost;

10 Year TCO Data Center New Mexico

Unfortunately, as the chart shows, the small reduction in operational energy cost simply cannot offset the first capital cost. Despite deliberately selecting a climate to favor the engineering decision there is no financial argument to support the additional capital expenditure.

Table 5 Real 10 year TCO of adiabatic option

Part two

Part two will look at the energy and environmental savings of the adiabatic option.

About the Author

Liam Newcombe

[executive name=”Liam Newcombe” function=”Guest Author” description=”Liam is a co-founder of Romonet and is recognised as a thought leader in the data center arena, Liam readily challenges convention to stimulate debate and collaboration to drive innovation. Liam has been a leading industry contributor to the European Commission’s European Code of Conduct for Data Centers (launched 19th Nov 2008), and used to chair the Best Practice Working Group. In 2011 Liam won the prestigious DatacenterDynamics ‘Outstanding Contribution to the Industry’ Award, the first non-American to do so.”]

To learn more about Romonet, visit

Share on FacebookTweet about this on TwitterShare on LinkedIn