Figure 2 describes a water-cooled system with the heat exchanger in series with the chiller, which is correct. Unfortunately, many systems are designed with the heat exchanger in parallel with the chiller, resulting in a much smaller number of winter hours of operation.
James B. Rishel PE
Dir of Mechanical Operations
Dear Mr. Rishel:
Mr. Rishel raises some interesting points. In my article I used 1000 kW as a simple, round number, but I could have used 17,500 kW with the corresponding savings for each system. All four free-cooling systems described in the article can be used up to the 5,000-ton system Mr. Rishel describes, but other factors come into play with larger installations.
Larger, water-cooled centrifugal compressors have a kW/ton of 0.5 to 0.6, while high-efficiency air-cooled screw compressor chillers using R-134a can now achieve 1.05 to 1.1 kW per ton. However, in most cases an evaporative cooling tower supplies cooling water to the condenser of a water-cooled chiller. Therefore the cost of the water-cooled chiller power must include the cooling tower fan motor(s), spray pump motors, and/or condenser water circulating pumps, plus the cost of makeup water, which closes or eliminates the gap between the two designs.
Air-cooled free cooling chillers occupy more outside space, while water cooled chillers occupy more expensive indoor space and require space for plate-and-frame heat exchangers, and piping and valves for a winter free-cooling system.
Water availability is a major risk factor for mission-critical cooling. A 5,000-ton chiller installation rejects approximately 6,700 tons of heat (with added power input from the compressors). A cooling tower requires about 2 gallons per minute (gpm) of makeup water per million Btu/h. As 6,700 tons = 80 million Btu/h, this amounts to 160 gpm (230,000 gallons/day) not including blow-down losses. Water treatment must be provided for this makeup volume.
As I mentioned in the article sidebar, a close study is needed to determine the best possible design for any large installation, considering total power cost, free cooling hours achieved, reliability, simplicity, and risk factors.
Figure 4 on p. 29 of the November/December issue was mislabeled. The figure shows a direct air-side economizer system.
Editor Fumbles on Good HandsDear Editor:
The chart on page 20 of the “Good Hands” article (Nov./Dec, p. 20) is flawed on two levels:
1- PUE should be measured as true power (kilowatts) not kilovolt-ampere (kVA).
2- Some of the cells seem to be missing info or mislabeled. This leads to errors such as the UPS output being greater than its input. Even assuming that the UPS input and output figures were inadvertently swapped, the calculated efficiency would be an inordinately optimistic ~99% (1251/1265 kVA or 1386/1396 kVA), which is not possible.
Current UPS technology usually runs 92-95 percent efficient.
Since this seems to be a 3-megawatt (MW) class data center, I found the lack of any mention of power distribution losses and (HV/MV/LV) transformer losses, as well any floor-level PDU distribution losses surprising. This lack of this information in the PUE calculation seems especially odd given the statement “a consolidated BMS, EMS, takes real-time measurements from all points all the way down to the cabinet level”.
I would have also like to have seen how much CRAH fan energy was used to meet the ~200 watt per square foot (sq ft) level that I calculated (2.66 MW/14,000 sf) (side note: were the CRAC fans powered by the UPS)? Were these perimeter downflow CRAHs? Was there any close-coupled cooling to used service the 18-kilowatt racks or was this all done using raised floor cooling?
What about other items like power for lighting 14,000 sf, or other items - (the 11 kVA for the office power panel seems out of place on this chart.
While I believe that the authors have designed an efficient data center, the article does not seem to really provide the key information to properly support their PUE claims or calculation methods.
Editor’s Note: An editing error caused much of the confusion pointed out by Mr. Neudorfer in point number 2. The table has been correctly reproduced on this page. The authors also noted that “kVA was used since the data were taken direct from the EPMS. The EPMS display is in kVA and since the calculations are based on actual verifiable data, kVA was used. The calculations do not switch between kW and kVA and use kVA throughout the analysis for consistency of data.”