1、2008 ASHRAE 17ABSTRACTTodays critical high-density electronic equipment envi-ronments need adequate equipment (rack) cooling withoutexcessive energy usage. This paper presents a methodology foroptimizing the rack cooling effectiveness and HVAC coolingcosts in telecom and data centers. Since raised-f
2、loor air distri-bution is the most common way of cooling data centers, thisscheme is used here to demonstrate the methodology. Two keydesign parameters are evaluated: The supply air temperatureand the supply airflow rate. These parameters not only impactthe rack cooling effectiveness but also the en
3、ergy costs for cool-ing the space. A leading Computational Fluid Dynamics(CFD) code is used to establish the temperature field andairflow pattern for various combinations of these parametersin a typical data center with hot and cold aisles. A new functionwas developed within the CFD code to compute
4、the Rack Cool-ing Index (RCI) for evaluating the thermal equipment environ-ment. This Index is designed to be a measure of how effectivelyequipment racks are cooled and maintained within industrythermal guidelines and standards. In addition, hypotheticalcost functions are introduced based on chiller
5、 and fan energycosts. Based on the RCI results and the energy cost functions,recommendations are given for optimizing the supply temper-ature and airflow. Specifically, the overall performance isimproved by modifying the temperature and airflow to valueshigher than traditionally thought useful. When
6、 the RCI algo-rithm has been incorporated into commercial CFD codes (orused manually), engineers and architects will have a new prac-tical tool to design and evaluate telecom and data centers foroptimal equipment rack cooling effectiveness and HVAC cool-ing costs.INTRODUCTIONRecent research suggest
7、that conventional under-floorcooling in data centers may have some inherent challenges inadequately cooling electronic equipment (Herrlin and Belady2006, Herrlin 2005, Sorell et al. 2005). Computational fluiddynamics (CFD) modeling compared this system with aconventional over-head system as well as
8、with a modular over-head solution. The three references theorize that the lack ofmixing in the cold aisle is one of the key reasons for theobserved difficulties for under-floor cooling to provide anadequate thermal environment. If correct, the selection ofsupply temperature and airflow rate should b
9、e critical. Amodeling study could help shed some light on the most appro-priate combination.Two prerequisites are necessary to proceed: (1) a measureof the rack cooling effectiveness and (2) a measure of the asso-ciated costs. The Rack Cooling Index (RCI) is a measure ofhow effectively equipment rac
10、ks are cooled and maintainedwithin industry thermal guidelines and standards (Herrlin2005). The Index is designed to help evaluate the equipmentroom “health” for managing existing environments or design-ing new ones. It is also well suited as a design specification fornew data centers. The Index was
11、 used in two of the referencesgiven above to evaluate the cooling effectiveness of over-headand under-floor air-distribution systems.In the present paper, different combinations of supplytemperature and airflow rate are analyzed for the impact on therack cooling effectiveness as expressed by the RCI
12、. For eachcombination, an established CFD code is used to determinethe airflow and temperature distributions in the entire datacenter, including the rack intake temperatures (Fluent 2006).Method for Optimizing EquipmentCooling Effectiveness and HVACCooling Costs in Telecom and Data CentersMagnus K.
13、Herrlin, PhD Kishor Khankari, PhDMember ASHRAE Member ASHRAE Magnus K. Herrlin is President of ANCIS Incorporated, San Francisco, CA. Kishor Khankari is CoolSim Product Manager at ANSYS, Inc.,Ann Arbor, MI.NY-08-0042008, American Society of Heating, Refrigerating and Air-Conditioning Engineers, Inc.
14、 (www.ashrae.org). Published in ASHRAE Transactions, Volume 114, Part 1. For personal use only. Additional reproduction, distribution, or transmission in either print or digital form is not permitted without ASHRAEs prior written permission.18 ASHRAE TransactionsThese data were subsequently analyzed
15、 by a new user functionin the code to compute the RCI.Cost functions are finally developed to assign the costs/savings of improving the RCI. In this particular case, the costfunctions are based on the main energy costs to cool the datacenter. More advanced functions can be developed to take firstcos
16、ts into consideration or ultimately a life-cycle approach.Although difficult, costs can also be assigned to the risk ofequipment failure at certain RCI levels. This first attempt,however, should highlight the potential of combining the RCIwith cost functions to provide comprehensive design informa-t
17、ion for the data center owner and/or consultant.RACK COOLING INDEX (RCI)The following is a brief overview of the Rack CoolingIndex (RCI) to provide the necessary understanding how tointerpret the Index. For a complete description of the RCI, thereader is referred to the original work by Herrlin (200
18、5) andpublished by ASHRAE.The Index deals with rack intake temperaturestheconditions that air-cooled equipment depend on for its contin-uous operation. The “allowable” equipment intake tempera-ture limits in Figure 1 represent the equipment test rangewhereas the “recommended” limits refer to target
19、facilityoperation. Over-temperature conditions exist once one ormore intake temperatures exceed the maximum recommendedtemperature. The total over-temperature represents a summa-tion of over-temperatures across all rack inlets. Similarly,under-temperature conditions exist when intake temperaturesdro
20、p below the minimum recommended. The numericalvalues of these limits depend on the applied guideline (e.g.,ASHRAE 2004) or de-facto standard (e.g., Telcordia 2001,2006). In other words, the RCI is a measure of the confor-mance with a given specification.The RCI has two parts, describing the equipmen
21、t roomhealth at the high (HI) end and at the low (LO) end of thetemperature range, respectively. Figure 1 provides a graphicalrepresentation of the RCIHI. An analogous Index is defined fortemperature conditions at the low end of the temperaturerange, RCILO.The RCIHIdefinition is as follows:(1)The ha
22、nds-on interpretation of the Index is as follows:RCIHI= 100% All intake temperatures max recommendedtemperatureRCIHImax recom-mended temperatureRCIHImax allow-able temperatureThe RCIHIis a measure of the absence of over-tempera-tures; 100% means that no over-temperatures exist (ideal).The lower the
23、percentage, the greater probability (risk) thatequipment experiences temperatures above the maximumallowable temperature. The Index for the hypothetical temper-ature distribution shown in Figure 1 is approximately RCIHI=95%. Based on numerous studies, a value at or above 95% isa sign of a good syste
24、m design.CFD analysis provides a convenient tool for computingand analyzing the RCI at various levels of detail. The RCI canbe calculated using all intake temperatures or any subsetthereof, down to a single equipment intake. While analyzedwith CFD modeling, the temperature for a single intake is the
25、average of all computational cells temperatures across thatopening. By studying the individual intakes RCI values, thevariation of the RCI across all air intakes can be determined.PARAMETER STUDYThe modeled 1400 ft2(300 m2) data center is shown inFigure 2. The equipment room has a 9 ft (2.7 m) ceili
26、ng and a2 ft (0.6 m) raised floor. There are 48 equipment racks12 perequipment line-upeach with 4 kW of heat dissipation. Thus,the total space load is 192 kW or 137 W/ft2(1475 W/m2). Thecooling airflow entering each rack is 640 cfm (1090 m3/h) withtotal rack airflow of 30,720 cfm (52,190 m3/h).Figur
27、e 1 Intake temperature distribution and graphicalrepresentation of RCIHI.RCIHI1Total Over-TempMax Allowable Over-Temp-100%=Figure 2 Data center layout showing electronic equipmentand CRAC units. Equipment air intake tempera-ture distribution at SAT = 55F (13C) and SAQ =80%.ASHRAE Transactions 19Ther
28、e are four computer room air conditioning (CRAC)units with a capacity of 17 ton (60 kW) cooling each. Thesupply temperature (SAT) is 55F (13C), 60F (16C),65F (18C), or 70F (21C). The supply airflow rate (SAQ)is 80, 100, 120, or 140% of the total rack airflow rate. Forty-eight 25% (open area) perfora
29、ted floor tiles deliver air fromthe raised-floor plenum into the cold aisle in front of theequipment.In the present paper, ASHRAE (2004) “Class 1” datacenter environment is used for the RCI calculations. In thisspecification, the recommended equipment intake tempera-ture range is 68 to 77F (20 to 25
30、C) and the allowable rangeis 59 to 90F (15 to 32C).COST FUNCTIONSUltimately, the costs/savings associated with improvingthe RCI need to be known. In this section, simple cost func-tions are developed to assign costs/savings based on the mainenergy costs to cool the space. More advanced functions cou
31、ldbe developed in a similar manner, taking first costs intoconsideration or deploying a life-cycle approach. Althoughdifficult, costs could also be assigned to the risk of electronicequipment failure. Nevertheless, this first attempt should high-light the potential of combining the RCI with cost fun
32、ctions toprovide comprehensive design information for the data centerowner and/or consultant.Energy costs for conditioning and distributing the air aremainly associated with chiller and fan energy, respectively.Increasing the supply air temperature allows operating thechiller at a higher evaporator
33、temperature andin turnhigher chiller efficiency and Coefficient of Performance(COP). Increasing the supply airflow rate, on the other hand,requires more fan energy to move air between the air-handlerand the data center. Clearly, both cost functions need to beevaluated when the supply air temperature
34、 and supply airfloware changed.For the modeled data center in Figure 2, four typicalCRAC units remove the equipment heat dissipation of192 kW. The COP was estimated at selected supply temper-atures between 55F (13C) and 70F (21C) based onmanufacturers data. Figure 3 shows the resulting chiller costf
35、unction, not including the evaporator fan energy. The annualenergy cost assumes an energy price of $0.08/kWh. Theseassumptions may or may not be applicable to a specificapplication or geographic location. Specifically, the chillercost function is a bit more complicated since the chiller effi-ciency
36、also is a function of the CRAC airflow. For clarity,however, this effect is not included to demonstrate the use ofcost functions.Eight fans (two per CRAC unit) transport the airbetween the units and the data center. Figure 4 shows theresulting annual fan cost as a function of airflow. One-hundred pe
37、rcent represents 30,720 cfm (52,190 m3/h) perthe discussion above. A total pressure drop of 1.5 in of water(373 Pa) was assumed as well as a fan efficiency of 75%.Again, these values may or may not apply to a specific appli-cation or geographic location.Figure 3 indicates that a supply temperature r
38、eduction of5F (3C) results in a cost penalty of $1600 whereas Figure 4points out that an airflow rate increase of 10 percentage pointsadds $500. These two hypothetical cost functions suggest thata supply temperature change has a greater impact on the oper-ating costs than does an airflow rate change
39、. Having such costfunctions at hand, a designer can evaluate different designswith regard to both RCI and operating costs.When the “sweet spot” has been established based on theRCI and costs, say, a supply temperature of 65F (18C) andan airflow of 120%, it can be used as the starting point whenmodif
40、ying the cooling design to further improve the RCI.Essentially, two design parameters have been fixed and thusremoved from the design process.RESULTSTables 1 and 2 show the RCI values for sixteen parametercombinations. Supplying an airflow rate that matches the rackairflow rate (100%) and supplying
41、the air at a “customary”55F (13C) temperature produce excellent thermal condi-tions at the high end of the temperature range (RCIHI)of100%) but very poor conditions at the lower end (RCILOof Figure 3 Hypothetical cost functionchiller.Figure 4 Hypothetical cost functionfan.20 ASHRAE Transactions20%).
42、 Indeed, numerous intakes draw air below the minimumallowable temperature of 59F (15C). Generally, this is notacceptable since it may be outside the manufacturers specifi-cations. Note that an “*” indicates that one or several intaketemperatures are outside the allowable temperature range.How can th
43、e performance of the raised-floor system beimproved? If both the RCIHIand RCILOare consideredequally important (if a range is specified, both should matter),a high supply flow rate combined with a high supply airtemperature should be evaluated. If the RCILOis of little inter-est, on the other hand,
44、a combination of 65F (18C) supplytemperature and 100% airflow may provide an adequate envi-ronment. A lower supply temperature or higher airflow willnot significantly improve the thermal conditions.By incorporating the proposed cost functions, the datacenter owner and/or consultant can assign a doll
45、ar amount toimproving the RCI values. Increasing the “customary” 55F(13C) supply temperature to 70F (21C) would result inbetter chiller efficiency. On an annual basis, this would saveabout 17% of the chiller energy or $4800. Note however, thathigh return temperatures may cause problems for many CRAC
46、units. Increasing the airflow rate from 100 to 140% on theother hand, would increase the fan operating costs by 40% or$2200. Increased airflow rates may exceed the availablecapacity of the raised floor plenum and the perforated tiles.These calculations assume unchanged system pressure dropby design.
47、Table 3 shows the net effect in percent of total annualchiller and fan energy, with the 55F (13C) and 100% as an(arbitrary) reference. As can be seen, the combination ofsupply temperature and airflow that provides ideal intakeconditions 70F (21C) and 120% also provides an energycost saving of 11%, a
48、 true win-win situation.The RCIHIand RCILOnumbers shown in Tables 1 and 2are based on all rack intakes temperatures in the data center.A visual representation of the RCI distribution is sometimesdesirable, including maximum and minimum values. Suchdistributions are shown in Figures 5 and 6 as comput
49、er-generated RCI maps for two parameter combinations: 55F(13C)/80% (base case) and 65F (18C)/100%. Since thefigures have the same RCI scale (0% to 100%) it is easy tocompare the effect of changes to the supply temperature and/or supply airflow rate.The figures highlight the substantial improvement of thethermal equipment environment by modifying the supplytemperature and the airflow rate. In this case, the combinationof 65F (18C) supply temperature and 100% flow
