1.1 kW/m2(100W/ft2) internal IT load.The cooling and air-handling systems for all three scenarios matchthe respective cooling and air-handling systems of the mid-tierdata center, except thatmore air-handling units are used tomeet anincreased internal heat load and greater cooling redundancyrequirements (2N). The Baseline scenario utilizes 17 CRAH unitsplaced on the data center floor. The Economizer and Economizer Plusscenarios use 10 air-handling units placed outdoors with on-boardair-side economizers. Two water-cooled (rather than air-cooled)chillerswith a cooling tower systemare assumed to be deployed forthis larger space type in all cases. In the two economizer scenarios,the economizer system is assumed to be custom designed for theenterprise data center, resulting in a lower supply-fan pressureresistance within the ducted air-handling system.2.6. Additional non-IT power loadsAlong with the mechanical systems, other significant non-ITenergy demands include lighting energy and UPS energy losses.Lighting energy density is assumed to be 10W/m2, whichapproximates observed lighting practices in conventional datacenters [19] and has been used to represent typical lighting energydensity in previous data center energy audits [12,13]. The lightingenergy density is assumed to be constant for all three scenariosacross all data center space types. Losses from UPS systems areestimated from empirical data of UPS efficiency relative to loadfactor [14], which indicates an efficiency of approximately 85% forany load factor greater than 0.2. The UPS load factor is calculated byfirst selecting UPS modules that provide adequate redundancyspecific to each space type and then determining the ratio of IT loadof total UPS capacity. The additional internal load from the lightingand UPS losses are added to the overall heat load that requiresmechanical cooling in the model.3. Results and discussionFig. 1 compares modeled data center energy use, expressed interms of PUE, for the Baseline, Economizer, and Economizer Plusscenarios, in relation to data center space type and location. TheBaseline PUE values are consistently less than 2.0 for all spacetypes and locations. The current stock of data centers is commonlyassumed by data center industry analysts to have a nationwideaverage PUE of approximately 2.0 [1,2]. This assumption is sup-ported by the small amount of available empirical data [14,20],byan industry consensus as reported in Brown et al. [1], and by recentindustry survey data gathered by the US Environmental ProtectionAgency [21]. Part of the discrepancy between the consensusunderstanding that the average PUE is ∼2.0 and the modelingresults presented here may be attributable to inefficiencies inoperation and airflow management that are not captured in themodel. The model assumes a homogeneous interior space andcorrectly implemented control sequences.
It does not captureimproperly functioning equipment or improperly programmedsystems that may occur in the operation of real data centers. Casestudies have assessed energy savings from improved airflowmanagement [22]. The energy savings potential indicated in thesecase studies corresponds approximately to the difference betweenour Baseline results and the generally applied PUE value of 2.0.In comparing the Baseline PUE values between different datacenter space types, larger data centers show relatively lower fanand cooling energy demand as more efficient mechanical equip-ment is used and as cooling loads are shifted from smaller DX unitsto larger chiller (central plant) systems. The Baseline PUE valuesalso become increasingly uniform among cities as data center sizeincreases and are nearly identical for the enterprise data centers.Cooling energy needs are similar for different cities in the Baselinecase, which is expected for data centers with mechanical designsthat employminimal outside air. On the other hand, the PUE valuesin the Economizer and Economizer Plus scenarios are substantiallydependent on climate, since the temperature and humidity ofoutside air must be suitable to reduce energy demand for cooling.For example, in the Baseline case, non-IT energy use for an enter-prise data center in Dallas is only 5% greater than in Seattle but thisdifference increases to 30% for the Economizer Plus scenario.Fig. 1 shows that the Economizer scenario could substantiallyreduce the energy use associated with DX cooling in the serverrooms to an extent that varies among climate regions. A 31%reduction in DX cooling is observed in the Dallas climate whereasDX cooling is reduced by nearly 86% in Seattle. A portion of thecooling savings is lost to increased fan energy use with the Econ-omizer scenario, which results from the increased air resistance ofthe HVAC ducting, relative to the ductless CRAC systems used in theBaseline case. The use of an exhaust fan during active economizerperiods also increases fan energy and necessitates greater overallfan energy in climateswhere the economizers can be used formorehours per year. The increased fan energy offsetsmuch of the coolingsavings and yields only modest PUE improvements in warmerclimates.In the Baseline design, increased mechanical redundancyrequires greater HVAC energy use in localized data centers ascompared to server rooms. The use of economizers for localizeddata centers results in significant DX cooling energy savings in themilder climates of San Francisco and Seattle, but less so in Chicago,Dallas, and Richmond, which experience more hours of elevatedoutdoor temperature and humidity. As in the server room spacetype, fan energy in the Economizer scenario for localized datacenters increases owing to the addition of exhaust fans andincreased air resistance associated with the ducted air deliverysystem. The effect is smaller for localized data centers than forserver rooms because more efficient fans and motors are availablefor the larger air-handling equipment.PUE results for the localized data centers most notably showthe increased savings available as economizer use is combined withrelaxedtemperature and humidity restrictions. Themodel shows thatcooling savings increasewhen the supply and return air temperaturesetpoints are raised, allowing the economizers to operate for more hours of the year. Additional savings are observed with the Econo-mizer Plus temperature setpoints, where the supply air temperatureremains at the Baseline setting and the return air temperature set-point is raised. The increased difference between the supply andreturn air temperatures allows for a reduction in supply airflow rate,resulting in less fan energy use. Under these Economizer Plustemperature settings, chiller activity is greater than it would be inresponse to simply increasing both the supply and return air set-points, but the reduced fan energy produces a net increase in energysavings.The shift from DX cooling systems in the server rooms andlocalized data centers to the use of air-cooled chillers inmid-tier datacenters results in less variation of the Baseline PUE values amongdifferent cities. The chillers in mid-tier data centers improve coolingefficiency, but no additional savings are realized in some regions,such as Seattle and Chicago, since the larger chiller systemsmatch theload size less accurately than the smaller DX units. As in the localizeddata centers, the increased chiller savings gained from economizeruse in the Economizer scenario for mid-tier data centers are partiallylost to the increase in fan energy required for the ducted air deliverysystem. While the air-handlers are more efficient in this larger datacenter, 建筑与环境空调英文文献和中文翻译(4):http://www.youerw.com/fanyi/lunwen_32446.html