1、22 2008 ASHRAEABSTRACTPlanning the information technology (IT) installation ina data processing environment is a cumbersome task. Eventhough Thermal Guidelines for Data Processing Environments(ASHRAE 2004) addresses heat load planning, the electricaland thermal considerations go hand-in-hand. Using
2、thecorrect electrical planning values can make the differencebetween either having adequate capacity or running out offinite power. The question becomes, what is correct? Thispaper will examine terms like nameplate and power consump-tion with an explanation of how each term should be appliedin the d
3、ata processing environment. In addition, a minimum setof electrical and thermal specifications are introduced.Finally, strategies are provided on how to properly plan theelectrical environment in tandem with the thermal environmentfor maximum energy efficiency.INTRODUCTIONThe data provided by manufa
4、cturers of equipmentdestined for the data processing environment is important tounderstand, not only during the purchasing cycle, but alsoprior to installation and while operating to adequately provi-sion power and cooling. However, understanding the data isoften confusing. The electrical and therma
5、l terminologyprinted in specifications must be read carefully and then prop-erly applied by the facilities team. As easy as planning theinstallation sounds, the specifications may contain too muchor too little information. If the specifications contain too muchinformation, making the correct choice
6、of electrical and heatload values is difficult, especially if all the data is scattered ina document. If the specifications contain too little information,educated guesses that depend on rules of thumb or past expe-riences are applied, but not always reliable. As data processingenvironments become m
7、ore energy conscious, rightsizing theenvironment by applying the specifications accurately beforethe information technology (IT) equipment enters intoproduction can have a big impact on energy efficiency andcapacity allocation. Interpreting specifications is more criticalthan ever before, but the va
8、riety of terms used must beexplained. Definitions of common electrical and thermalterms are presented in this paper, along with their intended useand possible misuse in the data processing environment. Auniform set of terms is suggested in a format that allows thefacilities team to quickly digest th
9、e electrical and thermalspecifications so that data processing resources are utilizedefficiently and effectively.As IT power and cooling needs tax the infrastructure, dataprocessing environments are reaching their design limitssooner than expected. Analyst groups like Gartner predict thathalf of the
10、 data centers in the world will have insufficientcapacity by the end of 2008 (Morgan 2006). The definition ofthe phrase “insufficient capacity” is different from one dataprocessing environment to the next as the design or operatingstrategies may include derating and/or redundancy (e.g., n+1)in the e
11、lectrical and mechanical systems. A typical deratingfor the electrical and mechanical systems is 80%-90% of ratedcapacity based on compliance to requirements like theNational Electrical Code, reliability, or guard banding. Dualredundancy (e.g., 2n) is used in many installations to increaseavailabili
12、ty, but this decreases capacity by at least 50% in anormal operating mode. Possible implementations for theelectrical distribution are shown in Figure 1.Option 1 is the most popular arrangement with two inde-pendent electricity paths (2n) that originate at or before theElectrical and Heat Load Plann
13、ingKeep Your Data ProcessingEnvironment RunningJoseph F. Prisco, PE Pamela L. Lembke Marvin M. MisgenJoseph F. Prisco is a senior engineer, Pamela L. Lembke is an engineer, and Marvin M. Misgen is an advisory engineer with IBM,Rochester, MN.NY-08-0052008, American Society of Heating, Refrigerating a
14、nd Air-Conditioning Engineers, Inc. (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.ASHRAE Transactions 23ma
15、ins service entrance. Both electricity paths remain indepen-dent up to the IT equipment. Since each electricity path is shar-ing the load through an Uninterruptible Power Supply (UPS),the UPS is loaded to a maximum of 50% under normal oper-ating conditions. If a derating factor of 80% is included, t
16、heUPS should not exceed 40% of its rated capacity unless a faultcondition exists in the data processing environment. Not onlyis capacity limited, but also studies by LBNL show that UPSefficiency is reduced as the percent of loading decreases (Ton2005). The same condition exists for the mechanical in
17、fra-structure. Efficiency for components such as chillers, coolingtowers, and other heat rejection units declines as the mechan-ical plant load moves further away from the maximum effi-ciency design point. Besides derating and redundancy, thereare other possible capacity limitations such as the main
18、 trans-former supplied by the utility or the lack of pole positions ina switchboard. Even though the design point ultimately limitsthe capacity, many data processing environments are exceed-ing their operating design point and are more at risk of notmeeting uptime objectives during planned maintenan
19、ce orunexpected outages.With finite capacity, thought must go into planning theelectrical and mechanical loads in the data processing envi-ronment. Manufacturers publish an assortment of values, butsome common terms are nameplate, maximum measuredpower consumption, and actual power consumption. Thei
20、rintended applications are as follows:Nameplate. The nameplate or power rating label con-tains the rated information, such as voltage, amperage,frequency, and phase. It should be used to size portionsof the electrical service (circuit breakers, branch cir-cuits, receptacle type) based on legally man
21、dated codesenforced by an authority having jurisdiction. The valueson the nameplate are derived in part by requirements inthe 60950-1 Information technology equipment - Safetystandard (IEC 2001) and through appropriate testing.The nameplate should always be higher than the maxi-mum measured or actua
22、l power consumption; however,IEC 60950-1 allows the equipment to consume up to10% more current than that stated on the nameplateunder normal operating conditions. This allows marginfor abnormally high loads and future hardware upgradesthat may consume more power. The nameplate has toalso take into c
23、onsideration the limitations of wall plugsand machine coupler ratings that may be critical to aspecific installation environment.Maximum measured power consumption. The maxi-mum measured power consumption is a value publishedby virtually all manufacturers. The value is based on acombination of empir
24、ically gathered test data and pre-dictions, or more typically, consists of measured valueson a fully configured piece of IT equipment with a veryhigh utilization factor. The value accounts for worst caseconditions, which may include a high ambient operatingtemperature, compensation for internal fail
25、ures that per-mit the IT to remain fully functional, and disablement ofpower aware technologies. The maximum measuredpower consumption should be used to size the electricaldevices which the nameplate does not cover.Actual power consumption. The actual power consump-tion is the most representative va
26、lue for a given configu-Figure 1 Electrical distribution options for equipment in the data center.24 ASHRAE Transactionsration. The value is available through manufacturer toolslike power calculators or a Thermal Report (ASHRAE2004). The value is an estimate of likely power con-sumption, but is infl
27、uenced by component and subsys-tem tolerances, application, as well as utilization. Actualpower consumption gives a reasonable heat load plan-ning number to use when sizing air conditioning insteadof nameplate or maximum measured power consump-tion.All three values, the nameplate, maximum measuredpo
28、wer consumption, and actual power consumption, must beused to initially plan the installation before the IT equipmentis up and running. The three terms require more explanationbeyond the definitions to avoid potential misuse.The biggest inhibitor to energy efficiency is the overprovisioning of the e
29、lectrical and mechanical infrastructure.Using the nameplate or a fixed percentage of the nameplatecan lead to over subscription. The maximum measured powerconsumption is useful for sizing the electrical infrastructureoutside of electrical code requirements that call for the use ofnameplate ratings.
30、The nameplate should not be substitutedfor the maximum measured or actual power consumption. For example, the National Electrical Code (NFPA 2005)defines loads as continuous and noncontinuous, where acontinuous load operates more than three hours at maximumcurrent. If a switchboard derated ampacity
31、is 200 amps and thenameplate on a piece of equipment is 20 amps, only ten piecesof equipment would be allowed on the switchboard. However,if the maximum measured amperage load is five amps, 40pieces of equipment would be allowed on the switchboard. Ifmaximum measured power consumption is used instea
32、d ofnameplate, the amount of IT equipment supported from oneswitchboard quadruples. The other possibility is to use the manufacturer providedactual power consumption to size the electrical infrastructure.Although the maximum measured power consumption mightnot be attainable on a daily basis, the ele
33、ctrical devices mustbe able to withstand the few, but quite probable, bursts ofpower demanded by the IT equipment. If the devices are notsized to handle the maximum attainable load, an over currentprotection device will de-energize and disrupt the flow of elec-tricity, possibly resulting in IT equip
34、ment downtime. If ashort-circuit analysis, including a selective coordination studyof circuit breakers, is not performed, the power outage couldbe more widespread and costly (Arnold 2006). Adding moreelectrical infrastructure capacity is one possibility, but itusually requires some part of the elect
35、rical distribution to bedisconnected from the supply to prevent electric shock. This istypically major work that requires a planned shutdown of crit-ical operations.Another consideration with respect to the actual powerconsumption is that data processing environments aredynamic as end user needs to
36、support applications and storageare increasing. If an IT equipment configuration changemanagement process is not in place, upgrades will likelyincrease power consumption. The uncontrolled addition ofprocessors, memory, hard disks, etc. in a chassis brings thepower consumption closer to the maximum m
37、easured powerconsumption and the tripping point of an over current protec-tion device. Without the proper checks and balances, using theactual power consumption to size the electrical facility canhave dire consequences. However, the actual power consump-tion gives very specific heat release informat
38、ion that is usefulfor rightsizing the air handling. If too low a heat release isplanned in the data processing environment, hot spots couldresult. Conversely, if too high a heat release (e.g., nameplateor maximum measured power consumption) is used, the airhandling unit capacity will be over-sized a
39、nd the reducedreturn air temperature will have a negative impact on the airhandling unit efficiency.The air conditioning system is better equipped to managethe short bursts of maximum heat release and maintain a setpoint within the programmed tolerance. The air conditioningsystem is also more forgiv
40、ing than the electrical system. Ashortage of volumetric airflow or conditioned air may result ininlet temperatures slightly exceeding the manufacturer speci-fication, but it does not cause an immediate shutdown. Addi-tional air conditioning can also be installed in parallel as theneed arises with mi
41、nimal impact on the data processing envi-ronment.Now that nameplate, maximum measured powerconsumption, and actual power consumption have beenexplored, the units of measure and their relationships arepresented.Typical units of measure include watts (W), kilowatt hour(kWh), volt amps (VA), British th
42、ermal unit (BTU/hr) andpower factor (PF). Although it is useful to have each bit of data,there are relationships between the units of measure that makeit possible to calculate missing pieces. Watts, the most heavilyused unit, is a measure of power consumption, whereas kWh isa measure of energy consu
43、mption. Power consumption andelectricity consumption are often misapplied or interchanged.Watts divided by VA results in the power factor. Their relation-ship is based on the power triangle that is discussed in manybasic electrical engineering textbooks. Multiply watts by aconstant of 3.413 to arriv
44、e at BTU/hr.The knowledge required to plan an installation is attainedfrom documents that contain the terms and their units as previ-ously described. Manufacturers often publish a lot of electricaland thermal data that is useful, but it is often incomplete orhaphazardly placed in a document. Table 1
45、 shows the data thatshould be available in one location. All the electrical and ther-mal planning data should be accessible from a single pagewhere the specifications are either published or a clear pointeris used to reference the location of the fundamental information.Although terms like nameplate
46、, maximum measuredpower consumption, and actual power consumption have aspecific purpose, in reality, each data processing environmentthat struggles with design point capacity limitations does notASHRAE Transactions 25use them exactly as defined. Rightsizing the data processingenvironment based on t
47、he actual behavior of the loads in agiven facility is important for energy efficiency. The informa-tion in Table 1 should be used to initially plan the data process-ing environment; however there are five practical strategies forensuring the longevity and energy efficiency of the hardwareinstallatio
48、n after the IT equipment is on-site and ready forconnection. The five strategies, documented below, are depen-dant on having trustworthy installation data over a period oftime. In the absence of installation data, planning the dataprocessing environment must revert to the guidance suggestedearlier i
49、n the paper.1. A pre-production test should be run to simulate the equip-ments operational behavior and to characterize thepower consumption. If possible, all server, storage,network, and support equipment should be tested. Manyfacilities teams do not know the actual power consump-tion of their IT equipment (Scaramella 2007). A pre-production test is one way to collect this vital data. Theresults equate to the actual power consumption definedearlier in this paper and can be used to size the air condi-tioning. Software tools like computational fluid dynam-ics (CFD) and engineering assessme
