By: Liam Newcombe
As promised in part 1, this second part of the post examines the energy and environmental savings of the adiabatic cooling option tested for TCO in part 1 and compares it with some other possible choices. To evaluate the environmental credentials we will look at:

·What happens to the TCO if we spend the $500k on renewable energy from solar panels instead?
·Does our data center use more or less water overall with the adiabatic option?

To recap, we are examining the impact of one specific option, which appears to be an obvious choice, for one specific design of data center in New Mexico. The data center is a modern 1.2MW design which uses direct outside air extensively for cooling, falling back to DX (direct expansion) mechanical cooling when the outside air is too dry, too wet or too hot for the IT supply conditions requested. For those who think the design looks rather like a Compass data center, you are right; the designs we are using are based on our model of the Compass design and the corresponding TCO outcomes apply.

The question in part 1 was whether it was worth the additional $500k capital cost to add adiabatic cooling to the direct outside air system to allow for vastly greater “economiser hours”. The outcome was, despite appearances, definitely not. Compass is not first saving capital cost at the expense of the customer TCO (although this is a recognised issue in the industry and a problem affecting many existing data centers); what Compass is providing is the smart TCO choice for their customer.

What about solar panels?
Having discounted the use of adiabatic coolers what would be the impact if, having shaved $500k off our capital budget, we spent it on solar photovoltaic renewable energy instead?
To assess this we need some data about the solar panels and how much sunlight we can expect through the year.

·Solar panel installations are frequently quoted as $ per peak kW, what this means is that our capital cost for installation gives us a peak kW capacity when the sun is overhead on a clear day, for this analysis I have used a cost of $2,000 per kW peak (kWP)[1], using our full $500k this allows for a peak solar capacity of 250kW.

·We won’t get 250kW for much of the time, certainly not overnight when it is dark, we will get less in the morning and evening and less when it is cloudy (which it isn’t very often in New Mexico). Fortunately the same climate data we used for the cooling performance analysis carries a typical year of solar radiation which we can use for a decent approximation of the annual energy output of the PV farm.
As before, we are considering both the 58F and 75F (contained airflow) supply temperature control options for the data center, each of which is considered in dry and adiabatic outside air variants. The chart below shows the daily total kWh demand from our previous analysis for each of the four options. In addition the available daily energy output of the 250kWP solar panel farm is shown to allow for comparison, in the best case the contribution is ~ 4.5%.

blog fig1

There is a common saying in the UK that something is “about as much use as a chocolate teapot” (due to the obvious thermal stability problem of chocolate to hot tea); in this case we might consider updating it to “about as much use as a solar panel on a data center”.

Financial assessment of Solar PV

Whilst the contribution to the overall energy consumption of the data center is small we should also assess the solar PV for TCO contribution as it does directly reduce electricity costs to the site. To do this we must make some assumptions about how the cost of electricity is going to change over the time we operate our investment. Most predictions for the US show power cost staying low for some time due to the influence of shale gas and other market forces, however environmental or energy security pressures may cause the value of solar generated power to increase over this time frame.

One of the oddities for this investment is that in many cases the operator would be better off locating the solar panels in an area with a feed-in tariff structure and selling the produced power to the grid as this would yield a greater income than the savings on energy achieved with the panels collocated with the data center and no available surplus to feed back.

To get a simple estimate of the value we keep energy costs flat and the same 7% discount rate we used in the original TCO assessment. Looking out over a full 20 years (investments in energy generation are not short term):

blog fig2

 

As shown in the table, the ROI is better than the adiabatic coolers but still not positive. To break even at the 20 year point we would need the value of the power generated by the solar panels to increase, specifically an annual increase in value of 0.9% less than our investment discount rate, or 6.9%. This is possible with legislative changes such as feed-in tariffs or a government applied cost of carbon but it is questionable whether this investment analysis should be part of the data center or a separate undertaking.

Water Consumption
An area of growing concern for data centers is water consumption; this has a direct (and likely to rise) cost impact which was dealt with in the TCO analysis. The secondary impact is the associated environmental impact of water consumption, particularly in areas of increasing water scarcity which includes many parts of the USA. To properly assess the utility of using water for “free” cooling of the data center we should consider whether there is a net increase or reduction in the water consumption associated with the data center.

This analysis is not as simple as it seems, we cannot usefully consider the water consumption of the data center without including the water used in the energy supply. Specifically, the WUE often quoted in marketing and “green dashboards” is of little relevance without understanding the water consumption involved in the electricity supply to the data center. This is included in the WUEsource variant of the metric which is provides a mechanism for us to compare the overall water impact. Obviously our data center uses more water than a dry cooled data center but substantially less than a data center which uses cooling towers. The question is; does it use less overall?

We have an analysis of the water requirement of the adiabatic system along with an estimate that 1.5 times this much water is actually used from the supply. The table in the Green Grid WUE[2] paper gives an Energy Water Intensity Factor (EWIF) of 2.38 litres per kWh for New Mexico. Applying this to the daily energy consumption we can derive the total water consumption and WUEsource for the four data center options, this is shown in Figure 2 below:

blog fig 3

It is clear from Figure 2 that there is no substantial difference between the options, indeed the order of preference varies from summer to winter as water consumption in the adiabatic humidifiers is traded with water consumption at the power station for mechanical cooling.

blog fig 4

Figure 3 shows how this adds up over a typical year, and we can see that the water consumption of the adiabatic options is higher overall. This is perhaps not surprising as it follows from the same reason that the adiabatic cooling doesn’t save money. The DX mechanical cooling provides anything up to 7kW of cooling for every 1kW of power consumed, this means that the adiabatic cooling on the data center would need to be 3.5 times[3] more efficient than the cooling towers at the power station to even keep up.

This outcome underlines the necessity to compare like with like and use realistic numbers when evaluating options for a data center. We have been generous to the adiabatic plant, in many cases the water treatment and losses could be 2x or more the actual adiabatic requirement, this analysis has used an estimate of 1.5 and still the adiabatic option came out worse. Other locations will have different water intensities of power generation, many lower. It is also not uncommon to see the case study for this type of device made by comparing the proposed system with an unrealistically poor baseline using unnecessarily restrictive controls and neglecting the substantial improvements in mechanical cooling efficiency delivered in recent years.

Comparing overall energy and water consumption

Unfortunately there is still no useful method of comparing water with source energy in most cases as the relationship between water and source energy is both complex and location dependent. In an area such as the Pacific North West USA or Northern Europe water is easily available at very low energy and environmental impact, in the South Western USA water availability is becoming a serious problem. This means that we cannot sensibly try to compare the WUE of data centers unless they have the same PUE as we do not know how much WUE it would be worth trading for a change of, say 0.1 in the PUE.

Whilst running the analyses and compiling the output for these two posts many of the results have surprised us both in their direction and magnitude. Hopefully these examples have provided a good example of why simple rules of thumb or free cooling hours estimates are not useful for assessing data center performance and that in many cases the real behaviours can be quite counter-intuitive. In other words, before committing yourself to a major investment in a data center, no matter how straightforward the benefit it will provide appears to be, it is always worth closely modeling how it will impact your TCO.

[3] Thermoelectric generation based on steam turbines is generally ~ 33% efficient producing electricity rejecting the remaining 67% source energy as heat, so the power station must reject twice as much heat energy as it supplies in electricity to power our DX chillers.
[2] http://www.thegreengrid.org/~/media/WhitePapers/WUE

[1] Based on continued oversupply of solar PV to the market from China and resulting falling prices, see http://www.forbes.com/sites/peterdetwiler/2012/12/11/solars-steady-march/ for a US market summary

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 www.romonet.com

Share on FacebookTweet about this on TwitterShare on LinkedIn