- 上一篇：氟伐他汀抑制昆虫保幼激素合成英文文献和中文翻译
- 下一篇：注塑模具的设计与热分析英文文献和中文翻译
theadditional savings in the Economizer Plus design are distributedfairly equally among the three larger data center space types, owingto the significant PUE improvements achieved from reducedoperation of the less efficient fan systems in the local and mid-tierdata centers.Regional differences in data center operation, which arerevealed in the Economizer and Economizer Plus scenarios in Fig. 1,are even more pronounced for greenhouse-gas emissions than forenergy use. Along with climate differences, site location affects themix of primary energy used to generate electricity. Table 3 presentsdistributions of electricity generation sources for the five cities usedto represent the regional variation of national data center opera-tion. Table 3 also categorizes the regional electricity resource mixusing three tiers of equivalent fossil carbon emissions, which arebased on previously reported analyses that aim to represent the entire lifecycle of electricity production. The carbon intensity ofelectricity production is expressed in units of grams of carbonequivalent per kilowatt hour of electricity delivered (gC(e)/kWh).Average lifecycle emissions from coal- and natural gas-generatedelectricity are estimated to be 280 and 155 gC(e)/kWh, respectively,with CO2 release attributable to combustion of coal and natural gasbeing the dominant source of these emissions [23]. For thepurposes of this paper, nuclear power, hydroelectricity, and otherrenewables (wind and solar) are all estimated to provide electricitywith lifecycle emissions of approximately 5 gC(e)/kWh. This cate-gorization is based on more detailed data in which nuclear andsolar have been estimated to emit 6.5 and 6.0 gC(e)/kWh, respec-tively, owing to greenhouse gases emitted during the extraction,processing, and disposal of associated materials [24]. Hydropowerand wind have been estimated to have a greenhouse-gas emissionsintensity of 5.5 and 1.4 gC(e)/kWh, respectively, accounting foremissions from infrastructure construction, flooded biomass decayin the reservoir, loss of net ecosystemproduction, and land use [25].All four of these electricity sources have markedly lower green-house-gas emissions intensities than either of the dominant fossil-fuel-based electricity generation technologies.In aggregate, there ismore than an order ofmagnitude differenceamong the estimated regional-average greenhouse-gas intensities,with Seattle producing the lowest emissions, 9 gC(e)/kWh, andDallasgenerating the largest at 188 gC(e)/kWh. The average greenhouse-gasemissions intensity of these five cities, 116 gC(e)/kWh, is lower thanthe estimatedUSnational average for electricitygeneration of160 gC/kWh, mainly because of Seattle’s and San Francisco’smuchsmallerdependence on coal-fired electricity relative to the nation as awhole.The wide range of greenhouse-gas intensities among the fiveevaluated cities highlights the potential impact of location whenaccounting for C(e) emissions associated with data centers. Thesavings of approximately 50% in non-IT energy (see Fig. 1)whenshifting fromthe Baseline enterprise data center scenario in Dallas tothe Economizer Plus case in Seattle, becomes a 95% difference in C(e)emissions when applying the greenhouse-gas intensities fromTable3. Conversely, the 40% energy savings observed when moving fromthe Baseline localized data center scenario in San Francisco to theEconomizer Plus design in Richmond corresponds to a 20% increase inC(e) owing to the change in regional greenhouse-gas intensity.4. Limitations and future researchThe modeled PUE values used to estimate the energy demandfrom non-IT components of data centers provide insight into themechanical cooling energy demand and efficiency improvementpotential for different types of data centers in the United States.However, the modeling results are based on assumptions ofequipment efficiency and operation that may differ from practiceowing to the many different mechanical designs available to datacenters. Here are brief statements of specific limitations andopportunities for further research.4.1. Characterization of server closetsTable 1 indicates that server closets represent about 13% of all ITdata center energy use. Empirical data are needed to better char-acterize this portion of the data center industry. The small amountof empirical data currently available to support the industryconsensus baseline PUE of 2.0 does not include server closets[14,20], which are considerably different in size and mechanicaldesign than other data center space types. Furthermore, potentialnon-IT energy savings associated with closet data centers, whilepossibly limited, are not included in this analysis because of thechallenge of disaggregating this energy use from overall officebuilding energy. Actual PUE measurements are needed to establishaccurate estimates of typical server closet performance and toestimate the energy efficiency potential of these small but commondata center spaces.4.2. Space type distribution of IT energyThe distribution of IT energy use among the different data centerspace types affects estimates of potential energy efficiencyimprovements because of the variation across space types inmodeled PUE values. However information regarding the distri-bution of IT energy among space types is currently only available fora single year [4]. Further analysis of the current and evolving datacenter stock is needed to validate the distribution of computerservers and IT energy use among different data center space types.4.3. Regional distribution of data centersModeling results highlight how energy use in smaller datacenters for the Baseline scenario and in all data centers in theEconomizer and Economizer Plus scenarios vary with climate. Thefive modeled cities, however, represent only a fraction of thepotential climates that data centers may experience. A betterunderstanding of where data centers are actually located and therespective local climate conditions for these locationswould permitmore refined estimates of national data center energy use andefficiency improvement potential.4.4. Greenhouse-gas emissions from data centersThe regional emissions intensities incorporated in this analysisillustrate how widely the greenhouse-gas emissions from elec-tricity generation can vary. Estimates of greenhouse-gas emissionsfrom data centers could provide an understanding of how changesin this industry can contribute to climate-change mitigation goals,but accurate emissions modeling would require electricity powermix data specific to each data center location. Seasonal and time-of-day variations in electricity power mix should also be investi-gated, since data center energy demand will increasingly vary withoutdoor climate conditions as economizers are incorporated intodata center design. Complete characterization of the climate-change impact associated with data centers would also requireaccounting for the embodied energy of the data center buildings,building equipment, and IT equipment, an important considerationthat has not yet been well characterized.Table 3Greenhouse-gas emissions intensity (gC(e)/kWh) associated with regionally specificelectricity generation sources.aCoal(280 gC(e)/kWh)Natural gas(155 gC(e)/kWh)Nuclear/renewables(5 gC(e)/kWh)Regionalaverage(gC(e)/kWh)San Francisco, CA 2% 47% 51% 81Seattle, WA 1% 1% 99% 9Chicago, IL 48% 4% 48% 143Dallas, TX 39% 49% 11% 188Richmond, VA 51% 10% 39% 160aThe distribution of electricity generation sources is based on local municipalitydata for San Francisco [26] and Seattle [27] and state averages for Chicago, Dallas,and Richmond [28]. Lifecycle gC(e)/kWh intensity estimates associated with elec-tricity generation in the United States are compiled fromJaramillo et al. [23] for coaland natural gas, Fthenakis and Kim[24] for solar and nuclear, and Pacca and Horvath[25] for hydro and wind. 5. ConclusionThe rapid growth of data center services and the resultingincrease in electricity to provide those services highlight the needto pursue energy efficiency opportunities in this economic sector.Building space types defined by IDC [1,8,9] have been evaluated toaccount for differences in mechanical equipment and operations.Results indicate that widespread economizer implementationincluding relaxed temperature and humidity operating conditionscould reduce the overall demand relative to the 2008 estimates [4]by about 20e25%, equivalent to an annual energy efficiencyresource of ∼13e17 billion kWh in the United States. Variationbetween the 2.0 PUE uniformly applied in previous data centerenergy estimates and the Baseline