Data Center Energy Reduction Strategies

INTRODUCTION

Data center designers and operators have a choice of strategies that can result in significant reduction in chiller plant connected power and consequently energy consumption. These options involve relaxing some of the tight temperature ranges under which plants were traditionally designed and operated. This paper will go over two strategies:

  1. Raise the Leaving Chilled Water Temperature (LCHWT) to make use of reduced lift and hence lower the connected power consumption.
  2. Increase the Temperature Delta T ( Δ T) between Entering and Leaving Chilled Water Temperatures, (ECHWT, LCHWT) to make use of pump laws and hence reduce pump horsepower.

We will show through examples how these strategies, combined with proper aisle containment practices, will result in 10-14% reduction in data center energy consumption at the low end and 15-18% at the high end.

WHY DATA CENTERS

The primary objective in computer rooms is to deliver a cool air stream to the servers in order to avoid overheating and shutdown of the microprocessor chip. The intent should not be to supply 55 ˚ F air to the servers with a comfort cooling objective; this will result in oversized chiller plants and unnecessary data center energy consumption. Both ASHRAE TC9.9 guidelines and LBNL 1 publications allow up to 80 ˚ F at the server inlet. Studies by server manufacturers have shown that their equipment can accept higher inlet air temperatures without failure. Since most of the cooling in data centers is sensible heat with very little latent loads, raising chilled water temperature in the data center is a guaranteed means of reducing chiller energy consumption; this is accomplished by lowering chiller lift and avoiding unnecessary dehumidification during the sensible cooling process.

Designers paid attention to ASHRAE guidelines and thus were born Hot Aisle (HAC) – Cold Aisle (CAC) containment layouts. This type of room layout enhances the delivery of a suitable amount of cool air to the front of the servers and minimizes the wasteful mixing of hot and cold air streams in the aisles– a step in the right direction. Unfortunately, designers and operators continue to be weary of implementing two energy conservation measures (ECMs): The first is raising LCHWT and room air supply temperatures. The second is continuing to be “stuck” to narrow temperature Delta practices typically 10 to 12 ˚ F.

In the next examples, we will show the extent of energy savings for each ECM.

RAISING SUPPLY TEMPERATURES

Raising temperatures (water and air) in a data center environment has many benefits for plant power and energy reduction. Raising the leaving LCHWT is one strategy that will reduce chiller lift. Comfort commercial cooling practices tend to emphasize on lowering lift by varying the condenser water temperature with ambient temperatures while maintaining a constant evaporator temperature; hence reducing the head pressure on the condenser side. With data centers, chilled water can be distributed at higher LCHWT hence reducing the head pressure on the evaporator side and achieving the same lower lift effect. Higher LCHWT is achievable because of the sensible cooling nature in data centers as previously mentioned.

Other benefits of raising temperatures are summarized as follows: (1) Setting air temperatures higher means less demand to cool the air streams; this means lower fan energy. Raising the temperatures for either air or water increases the number of hours per year where (2) water side or (3) air side economizers may be used. The compressor is either off or at partial loading during full or partial free cooling and energy consumption is decreased. (4) In addition, smaller chiller power requirements may translate to smaller generator sizes that satisfy the electrical load of the building. In present-day energy “lingo”, all the above translate to (5) smaller Power Usage Effectiveness (PUEs) for data centers and (6) reduced carbon footprint for projects.

Sample

For purposes of this paper, a 1000 ton centrifugal chiller is selected for various LCHWT starting at 42 ˚ F and up to 60 ˚ F in increments of 1 ˚ F. The evaporator temperature differential is kept constant at 14 ˚ F Delta across the selections; similarly condenser water conditions are kept at a 12 ˚ F Δ T with 85 ˚ F leaving and 97 ˚ F entering condenser water conditions. Actual selections are provided by TRANE for each temperature. Figure 1 plots the primary chiller connected power in kilo Watts (kW) for each LCHWT. The results are as anticipated; the reduction in power is approximately 115 kW or 29% between the two extreme leaving water supply temperatures. Considering that most chiller plants operate in the 42-46 ˚ F at a constant delta, elevating the LCHWT up to the mid-50 ˚ F’s reduces power consumption by 14-18%. Taking a more aggressive approach and raising the LCHWT to the low 60 ˚ F’s results in significant power reductions in the range of 25-32%. With the proper CAC/HAC containment strategies and chiller controls, raising the supply air temperature in conjunction with the LCHWT is not only achievable, but proves to have significant benefits for data center power reduction.

Data Center Energy Consumption

WIDER TEMPERATURE DELTA

Another strategy to reduce power consumption is by increasing the Temperature Delta ( ΔT ) between LCHWT and ECHWT. Flow and Δ T are inversely related according to the formula

Data Center Power Management

For a constant cooling load, as D T increases, flow decreases linearly (1:1). However pump horsepower is related to the cube of flow as shown in the pump equation:

Data Center Energy Consumption

As flow decreases, pump horsepower decreases to the third power. This translates to a considerable reduction in pump horsepower and energy consumption for the overall plant. Subsequently, reduced flow and smaller pumps translate to smaller Variable Frequency Drives (VFD’s) and pipes throughout the facility.

Sample

For purposes of this paper, we simulated a 4000 ton chilled water plant serving a 100,000 sq.ft data center at 150 W/sf (Washington, DC). The plant is served by four (4) 1000 ton centrifugal chillers, four (4) 1000 ton induced draft cooling towers, and a variable primary pumping system. LCHWT’s were varied starting at 42 ˚ F up to 56 ˚ F in increments of 2 ˚ F.

Data Center Energy

Two extreme temperature deltas are selected; this provides a good picture to interpolate temperatures in between these deltas. The Base Plant (Table 1) is simulated at 10 ˚ F chilled water ∆ T. The alternate plant (Table 2) is simulated at 16 ˚ F ∆ T for the chilled water system. Both plants were simulated for 12 ˚ F ∆ T for the condenser water side with 85 ˚ F Leaving Condenser Water temperature. Energy costs were estimated using a utility rate of $0.08/kWh.

Data Center Consumption

The results for total energy consumption for both temperature deltas are charted in Figure 2, but an important observation needs to be highlighted first. Figure 2 plots the LCHWT’s and the corresponding Return Air (RA) Temperature at the CRAH2 units. RA Temperature is a snapshot of the overall room temperature, and what is important here is the direct correlation with LCHWT. Room air temperature increases proportionately with LCHWT and that will have a negative impact on a data center thermal environment. Simply put, there has to be proper HAC/CAC containment to route the hot air back to the CRAH units. If containment is not implemented, more CRAH units will be required to offset the higher return air temperatures and more energy will be consumed by CRAH unit fan motors than chillers and pumps. Containment is factored into our computer models and hence RA temperatures floated up with the LCHWT without additional fan horsepower at the CRAH units.

Energy Costs in the Data Center

Chilled water pump Brake Horsepower (BHP) is plotted in Figure 3 starting at 10 ˚ F ∆ T and up to 16 ˚ F ∆ T in increments of 1 ˚ F. BHP is reduced by 57% at the extreme temperature range points. This power reduction will in turn have a profound effect to reduce overall plant energy consumption as will be shown next.

Data Center Efficiency

Total plant energy consumptions for the BASE (10 ˚ F ∆T ) and the ALTERNATE (16 ˚ F ∆T ) are plotted in Figure 4. In concurrence with the first strategy, energy is reduced along a constant delta T selection (follow either the green or the red lines for any LCHWT). The reduction in energy between 44 ˚ F and 56 ˚ F is ± 13% for either the BASE or the ALTERNATE scenarios. The reduction in energy is an additional 6-8% as delta T is increased –a jump from the green to the red line along any of the LCHWT points.

Data Center Plant Energy

 

Data Center Cooling

Since most chiller plants are designed around a 44-45 ˚ F LCHWT with 10 ˚ F ∆ T, a user may want to know what percentage energy is reduced between any LCHWT at the 16 ˚ F ∆ T and the BASE Temperature (44 ˚ F, 10 ˚ F ∆ T). The results plotted in Figure 5 show a reduction in energy between 6% to 18%. A “Sweet Spot” is shown where LCHWT are in the low 50 ˚ F range; the energy reduction is 10%-14%. The term Sweet Spot is designated because these temperatures are reasonably easy to achieve without dramatically affecting either the chiller controls or the interior thermal conditions within the data center. Increasing LCHWT temperatures to the upper 50’s or low 60’s ˚ F results in further energy reduction between 15-18%. This practice, while achievable, should be closely coordinated with the chiller manufacturers in order to mitigate the risks of low lift conditions. Lower ∆ T lines, while not simulated, are sketched on the graph to give the user an idea of where the percentage reductions will in turn fall on the chart and hence interpolated.

CONCLUSION

Design engineers and operators can reap significant energy and cost savings when selecting proper chilled water set points for LCHWT and temperature deltas. Data centers are ideal candidates for LCHWT in the mid 50 ˚ F’s with high Temperature Delta. These combined strategies can reduce data center power and energy consumption from a moderate low of 10-14% up to an aggressive high between 15-18%.

REFERENCES

[1] ASHRAE TC 9.9
[2] “High Performance Data Centers”, LBNL and Pacific Gas and Electric Study, 2006
[3] “Chiller Design for Low-Lift Conditions”, Tony Doyon, AC/H/R The News. May 12, 2008
[4] “OA Economizers for Data Centers”, Vali Sorell, ASHRAE Journal, December 2007
[5] “Optimizing Chilled Water Plants”, David Kelly, HPAC, January 1999

1 LBNL: Lawrence Berkeley National Laboratories
2 Computer Room Air Handling Unit