Tooling to Improve Data Center Efficiency A Survey and Considerations for the SMB Data Center

原文作者：

Shannon McCumber，計(jì)算機(jī)系，威斯康辛大學(xué)帕克塞德分校，美國

Susan J Lincke，計(jì)算機(jī)系，威斯康辛大學(xué)帕克塞德分校，美國

原文發(fā)表于2014年IEEE學(xué)報(bào)

INTRODUCTION

前言

This is a survey paper of recent research conducted on the opportunity to implement tools to increase data center efficiency and reduce the demand for energy to run data centers. This is important since data centers account for approximately 2% of total U.S. electricity consumption and that this number has increased 65% since 2000 [25, 26]. Our concern is that a tool is available for smaller data centers, in addition to larger ones. These small centers need a lower cost solution that is easy to understand by a regular IT person.

這是一篇近期的調(diào)查論文，研究了如何利用工具提高數(shù)據(jù)中心效率，減少數(shù)據(jù)中心運(yùn)行能源需求的問題。這個(gè)問題很重要，因?yàn)閿?shù)據(jù)中心用電量約占美國總用電量的2%，而且這個(gè)數(shù)字自2000年以來已經(jīng)增長了65%。我們關(guān)注的是，相對于較大型數(shù)據(jù)中心，適用于較小型數(shù)據(jù)中心的工具。這些小型中心需要一種低成本的解決方案，且易于為普通IT從業(yè)人員所了解。

Genesh et al [2] propose an integrated approach to power management for the data center using two key concepts: ‘power proportional’ and ‘green data center’. These approaches focus on reducing the power consumption of disks and servers (power proportional) and the support infrastructure such as cooling equipment, power backups and distribution units (green data center). Our paper focuses on these two areas, by including a section for each area of improvement. In addition, we review comprehensive research tools, which implement a variety of techniques to green data centers.

Genesh等人利用“功率比例”和“綠色數(shù)據(jù)中心”這兩個(gè)關(guān)鍵概念，提出了數(shù)據(jù)中心電力管理的綜合方法。這些方法關(guān)注于減少磁盤和服務(wù)器的功耗(按功率比例)以及制冷設(shè)備、后備電源和配電單元(綠色數(shù)據(jù)中心)等支撐基礎(chǔ)設(shè)施。我們的論文關(guān)注于這兩個(gè)領(lǐng)域，包括每個(gè)領(lǐng)域的改進(jìn)部分。此外，我們回顧了應(yīng)用在不同技術(shù)綠色數(shù)據(jù)中心的綜合研究工具，

We begin by reviewing general purpose data center efficiency metrics, mentioned in these case studies and research. Power usage efficiency (PUE) is a measure of how efficiently a data center uses its power. PUE is the ratio of total facility energy to the energy consumed by computing equipment:

PUE = TotalEnergyToFacility / ComputePwr      (1)

 我們首先回顧了這些案例研究中提到的通用數(shù)據(jù)中心效率指標(biāo)。電能利用效率(PUE)是衡量數(shù)據(jù)中心使用電能效率的指標(biāo)。PUE為設(shè)備總能耗與計(jì)算設(shè)備能耗的比值:

PUE = 設(shè)備總能耗 / 計(jì)算設(shè)備能耗    (1)

PUE is the most common green metric used to evaluate the efficiency of a data center and in the ideal case this value approaches 1.0 [10]. Google has data centers that have measured PUE values as low as 1.09, and they stress its importance in optimizing data center efficiency [13]. A similar metric is the coefficient of performance (COP), which is the ratio of the heating or cooling provided over the electrical energy consumed.

PUE是評估數(shù)據(jù)中心效率的最常見的綠色指標(biāo)，在理想情況下PUE值接近1.0。谷歌的數(shù)據(jù)中心測量PUE值低至1.09，他們很重視優(yōu)化數(shù)據(jù)中心效率的重要性。一個(gè)類似的指標(biāo)是性能系數(shù)(COP)，它是提供的采暖或制冷與所消耗電能的比值

Energy Reuse Effectiveness (ERE) is a green data center metric, which represents the amount of energy (or wasted heat) that can be reused by, for example, heat-activated cooling. ERE is measured as:

 (TotalElec – EnergyReused) / ElecUsedByITequip (2)

再生能源利用效率(Energy Reuse efficiency, ERE)是一個(gè)綠色數(shù)據(jù)中心指標(biāo)，它代表了可以再利用的能源(或余熱)的數(shù)量，例如，熱激活冷卻。ERE表示為:

（ 總能耗 – 再利用能量) / 設(shè)備能耗    (2)

SERVER UTILIZATION AND POWER MANAGEMENT

服務(wù)器使用和電源管理

The U.S. Department of Energy [15] suggests that most

servers in the data center run at 20% or less utilization while drawing full power. Consolidating these underutilized resources is their ‘number one’ method to implement additional efficiency measures within existing data centers [14].

美國能源部指出,大多數(shù)數(shù)據(jù)中心中的服務(wù)器運(yùn)行利用率為滿負(fù)荷的20%或更少。整合這些未充分利用的資源是他們在現(xiàn)有數(shù)據(jù)中心內(nèi)實(shí)施額外效率措施的“首要”方法。

Reducing the number of devices in the data center contributes to the overall goal of data center efficiency. Rack servers consume the most IT energy in the average data center [15]. Virtualization allows multiple independent systems to run on a single physical server and can drastically reduce the number of servers in a data center. In a dedicated data center, one way to reduce the number of servers is to consolidate applications. Retiring legacy, redundant and underutilized applications will free up resources. In one example, application consolidation resulted in a 60% reduction in energy costs [14]. Additional energy savings can be realized by consolidating redundancies in the IT system and sharing resources such as: power supplies, CPUs, disk drives and memory [15].

減少數(shù)據(jù)中心中的設(shè)備數(shù)量有助于實(shí)現(xiàn)數(shù)據(jù)中心效率的總目標(biāo)。在普通數(shù)據(jù)中心中，機(jī)架服務(wù)器消耗了最多的IT能量。虛擬化技術(shù)允許在單個(gè)物理服務(wù)器上運(yùn)行多個(gè)獨(dú)立的系統(tǒng)，可以大大減少數(shù)據(jù)中心中的服務(wù)器數(shù)量。在專用數(shù)據(jù)中心中，合并應(yīng)用程序是減少服務(wù)器數(shù)量的一種方法。遺留、冗余和未充分利用的應(yīng)用程序?qū)⒈会尫刨Y源。在某個(gè)案例中，應(yīng)用程序整合降低了60%的能源成本。通過整合IT系統(tǒng)冗余并共享資源(如:電源、CPU、磁盤驅(qū)動器和內(nèi)存)，可以實(shí)現(xiàn)額外的節(jié)能。

Reducing the direct consumption of power by the data center is one of the most effective and least expensive approaches to Green IT [14]. Typical data center servers are built to manage peak loads while sitting idle the majority of the time. Idle servers consume about 60% of peak power [28]. A power proportional approach aims to focus the processing load onto a subset of servers to allow the remaining idle resources to be powered down. Ganesh et al [2] suggest powering down

減少數(shù)據(jù)中心的直接電力能耗是實(shí)現(xiàn)綠色I(xiàn)T的最有效和最便宜的方法之一。典型的數(shù)據(jù)中心服務(wù)器是為了管理峰值負(fù)載而構(gòu)建的，而這些負(fù)載在大多數(shù)時(shí)間處于閑置狀態(tài)。閑置服務(wù)器消耗了約60%的峰值功率。功率比例方法的目的是將運(yùn)算負(fù)載集中到一部分服務(wù)器上，以便關(guān)閉其余閑置資源。

“containerized data centers” which would include not only the servers but also the power distribution, backup, networking and cooling equipment. Another power management approach uses Energy Star or high-performance processors, which use performance states to control power consumption by varying the voltage of a component depending on the circumstances. Performance states which undervolt decrease the voltage used in a component thus saving energy. This technique was first introduced to CPUs, but can also be used to control the active cache size, the number and/or operating rates of memory and I/O interconnects [22].

“集裝箱數(shù)據(jù)中心”不僅包括服務(wù)器，還包括配電、備份、網(wǎng)絡(luò)和制冷設(shè)備。另一種功率管理方法使用能源之星或高性能處理器，它們利用性能狀態(tài)來控制功耗，方法是根據(jù)情況改變組件電壓。電壓不足的性能狀態(tài)會降低組件電壓從而節(jié)省能源。這種技術(shù)最初被用于CPU，但也可以用于控制在用緩存大小、內(nèi)存和I/O接口數(shù)量和/或運(yùn)行比例。

AIR FLOW

氣流

Since air conditioning is a major factor in data center inefficiency, optimizing air temperature and flow is important. Computational Fluid Dynamics (CFD) is a tool widely used during this effort. Blogs indicate that commercial tools price from $7500 (CoolSim) to $30,000 (Future Facilities 6Sigma).

由于空調(diào)是造成數(shù)據(jù)中心效率低下的主要因素，因此優(yōu)化空氣溫度和氣流至關(guān)重要。計(jì)算流體力學(xué)(CFD)軟件是一種廣泛用于解決該類問題的工具。博客文章指出，商業(yè)工具軟件的價(jià)格從7500美元(CoolSim軟件)到30000美元(Future Facilities 公司的6Sigma軟件)不等。

Model Air Temperature

空氣溫度模型

CFD tools are used to predict the flow field in a region of interest and the distribution of temperatures. CFD uses numerical methods and simulation to analyze air or fluid flow. A CFD model of the air flow in the current data center configuration is used as a baseline to compare against the results of modifications made. Servers, switches, UPSs, power distribution units (PDUs) and other components used in the data center each have a model. The flow fields that result from this simulation can be used to pinpoint hotspots, analyze new data center configurations and understand the distribution of workload to cooling equipment [23]. Google [13] used such data to design a solution using low cost materials such as meat locker curtains and sheet metal to contain hot and cold aisles.

CFD工具用于預(yù)測研究區(qū)域的流場和溫度分布。CFD使用數(shù)值方法和仿真來分析空氣或流體流動。將當(dāng)前數(shù)據(jù)中心配置中的氣流CFD模型作為基準(zhǔn)，與修改后的結(jié)果進(jìn)行比較。數(shù)據(jù)中心中的服務(wù)器、交換機(jī)、UPS、配電單元(PDU)和其他組件都有各自的模型。通過仿真得到的流場可以準(zhǔn)確定位熱點(diǎn)，分析新的數(shù)據(jù)中心配置，了解冷卻設(shè)備的工作負(fù)荷分布情況。谷歌利用這些數(shù)據(jù)設(shè)計(jì)了一種解決方案，使用低成本的材料，如冷庫隔簾和金屬板來封閉冷熱通道。

A lower cost alternative is to simply measure rack inlet and outlet temperatures. The Supply Heat Index (SHI) is a metric which describes this convective heat transfer across a rack [3]. SHI is calculated using the inlet and outlet rack temperatures (InletT, OutletT), and the supply temperature (SupplyT), which is the temperature delivered by the air conditioning unit:

 SHI = (InletT – SupplyT) / (OutletT – SuppyT)      (2)

 另一種低成本替代方法是直接測量機(jī)架進(jìn)風(fēng)和出風(fēng)溫度。送風(fēng)熱指數(shù)(SHI)是一個(gè)描述穿越機(jī)架的對流換熱情況的參數(shù)。SHI的計(jì)算采用機(jī)架進(jìn)出風(fēng)溫度(InletT, OutletT)和送風(fēng)溫度(SupplyT)，即空調(diào)機(jī)組提供的溫度:

SHI = (進(jìn)風(fēng)溫度-送風(fēng)溫度)/（出風(fēng)溫度-送風(fēng)溫度)  (2)

Sharma et al [3] conducted a set of experiments that measured temperature distribution across a production data center facility using SHI calculations. The results of these calculations showed that the SHI can successfully be used to understand heat transfer and fluid flow and reduce the overall energy consumption of data centers [3].

Sharma等人進(jìn)行了一組實(shí)驗(yàn)，使用SHI計(jì)算方法測量了生產(chǎn)數(shù)據(jù)中心設(shè)備上的溫度分布。這些計(jì)算結(jié)果表明，該方法可以成功地用于分析熱傳遞和流體流動，降低數(shù)據(jù)中心的整體能耗。

Breen et al [6] developed a model that showed that by increasing the air temperature supplied to each rack inlet resulted in potential gains for COP. The data center COP improves approximately 8% for every 5 degrees C increase. However the guidance for colder temperatures and low relative humidity have been relaxed since these factors have less impact on server performance than once thought [16].

 Breen等人開發(fā)了一個(gè)模型，該模型表明，通過提高每個(gè)機(jī)架入口的空氣溫度，可以為COP帶來潛在的收益。數(shù)據(jù)中心每升高5℃COP大約能提高8%。然而，對于較低溫度和較低相對濕度的限制已經(jīng)放寬，因?yàn)檫@些因素對服務(wù)器性能的影響比以前認(rèn)為的要小。

Hamann et al [4] used real time sensors to determine thermal zones within the data center. Air flow measurements were used to track air from different areas of the data center to the air conditioning units (ACU).

Hamann等人使用實(shí)時(shí)傳感器來確定數(shù)據(jù)中心內(nèi)的熱區(qū)域。氣流測量用于追蹤從數(shù)據(jù)中心不同區(qū)域到空調(diào)機(jī)組(ACU)的氣流。

Marshall and Bemis [23] propose that in addition to design and monitoring in the future, CFD will be used in real time analysis for data center energy efficiency. To extend this concept, a physics-based modeling approach has shown that it is possible to create the desired air flow patterns that are believed to help optimize the energy efficiency within the data center [4].

Marshall和Bemis提出，未來除了設(shè)計(jì)和監(jiān)測，CFD還將用于數(shù)據(jù)中心能效的實(shí)時(shí)分析。為了擴(kuò)展這個(gè)概念，基于物理的建模方法表明，可以創(chuàng)建所需的流型以優(yōu)化數(shù)據(jù)中心的能源效率。

Wider Temperature/Humidity Tolerances

更寬的溫/濕度允許范圍

Environmental control within the data center focuses on monitoring and controlling air temperature and humidity. Raising air temperature reduces energy use, but could lower reliability, availability, and equipment life expectancy. As a result of a set of experiments (including those described earlier), ASHRAE has widened the recommended range of temperatures within the data center to 18-27 degrees C for all four classes, with an allowable range of 5-45 degrees C for their class 4 of equipment. Their maximum recommended relative humidity (RH) is up to 60% with an allowable maximum of 80% RH [16].

數(shù)據(jù)中心的環(huán)境控制主要是對空氣溫濕度的監(jiān)測和控制。提高空氣溫度可以減少能源消耗，但可能降低可靠性、可用性和設(shè)備預(yù)期壽命。作為一系列實(shí)驗(yàn)的結(jié)果（包括前面已描述的），ASHRAE（美國采暖、制冷與空調(diào)工程師學(xué)會）已經(jīng)將數(shù)據(jù)中心4個(gè)級別的建議溫度范圍擴(kuò)大到18-27℃，4級設(shè)備的允許溫度范圍為5-45℃。它們的最大推薦相對濕度(RH)高達(dá)60%，允許的最大相對濕度為80%。

Breen et al [6] developed a model that described the “heat flow from the rack level to the cooling tower for an air cooled data center with chillers.” This model was then used to study the effects of varying the temperature supplied to the rack as well as the temperature rise across the rack. One case study extended the rack air inlet temperature (beyond ASHRAE recommendations) over the range 5-35 degrees C while holding the temperature across the rack constant. The second case held constant the rack air inlet temperature while varying the temperature across the rack from 5-35 degrees C. These studies [6] show that increasing either the rack air inlet temperature or the temperature across the rack both yield improved energy efficiency. The higher temperatures were acceptable, but the benefits gained by increasing the temperature rise across the racks outperformed the benefit of increased temperature at the air inlet.

Breen等人開發(fā)了一個(gè)模型用以描述“帶冷水機(jī)組的風(fēng)冷數(shù)據(jù)中心從機(jī)架級到冷卻塔的熱流體”。這個(gè)模型用來研究機(jī)架進(jìn)風(fēng)溫度的改變和穿越機(jī)架溫升的影響。某個(gè)案例研究中，將機(jī)架進(jìn)風(fēng)溫度(超出ASHRAE的建議)擴(kuò)展到5-35℃范圍，同時(shí)保持機(jī)架上的溫度恒定。另一個(gè)案例中保持機(jī)架進(jìn)風(fēng)溫度恒定，同時(shí)改變機(jī)架的溫度從5-35℃。研究表明，提高機(jī)架進(jìn)風(fēng)溫度或機(jī)架兩側(cè)溫度都能提高能效收益。較高的溫度是可以接受的，但通過提高機(jī)架的溫升所獲得的效益優(yōu)于提高進(jìn)風(fēng)溫度的效益。

FREE AIR COOLING

自然空氣冷卻 

Free Air Cooling employs air side economizers, which use outside air at a favorable temperature in lieu of chilling units to cool the air delivered to the servers. Hot air drawn away from the servers is then expelled outside instead of being recycled and chilled.

 自然空氣冷卻采用風(fēng)側(cè)經(jīng)濟(jì)器，它使用室外適合溫度的空氣，而不是冷水機(jī)組冷卻空氣送到服務(wù)器。從服務(wù)器抽走的熱空氣會被排出室外，而不是被回收和冷卻。

Facebook [19], Google [13] and Intel [17] data centers have all implemented this technology. Google uses free air cooling in all of its data centers, Facebook implemented free air cooling as part of a continued effort to achieve LEED gold certification and Intel found that its use of free air cooling 91% of the time instituted a 74% reduction of power consumed by the data center. Klein et al [24] agree that significant energy savings can be realized by implementing free air cooling. The use of sustainable energy sources to power the data center further drive green efficiencies and reduce the overall demand on non-renewable energy sources [18].

Facebook、谷歌和Intel數(shù)據(jù)中心都實(shí)現(xiàn)了這一技術(shù)。谷歌在其全部數(shù)據(jù)中心使用自然空氣冷卻；Facebook將自然空氣冷卻作為實(shí)現(xiàn)LEED金牌認(rèn)證持續(xù)努力的一部分；英特爾發(fā)現(xiàn)其數(shù)據(jù)中心91%的時(shí)間使用自然空氣冷卻導(dǎo)致能源消耗減少74%。Klein等人認(rèn)為實(shí)現(xiàn)自然空氣冷卻對節(jié)能具有重要意義。利用可持續(xù)能源推動數(shù)據(jù)中心進(jìn)一步提高綠色能效，減少了對不可再生能源的總體需求。

During their quest for Leadership in Energy & Environmental Design (LEED) Gold Certification, Facebook created the Open Compute project. “The Open Compute Project is a set of technologies that reduces energy consumption and cost, increases reliability and choice in the marketplace, and simplifies operations and maintenance.” [31] Facebook implemented evaporating cooling which draws in air from the outside, filters it, and cools it by spraying a mist of purified water into the air, cooling it as it evaporates. The cooled air is then passed through another series of filters to ensure that the air is of the correct temperature and humidity before being delivered to the data center. This filtered outside air system may be turned off if too hot, or used if in the allowable temperature range, or may be diluted with data center air, if the outside air is too cool. Initially, the air was cooled to 80.6 degrees F based on ASHRAE recommendations and Facebook was preparing to make a change to 85 degrees F based on the results of their implementation at the time the video was made. In addition, the Facebook data center uses LED lighting, rain water for flushing the toilets, and other such measures to further reduce the overall energy consumption by its data centers [19].

Facebook在尋求“能源與環(huán)境設(shè)計(jì)先鋒” (LEED)金級認(rèn)證期間，創(chuàng)建了開放計(jì)算項(xiàng)目。開放計(jì)算項(xiàng)目是一系列降低能源消耗和成本、增加可靠性和市場選擇、簡化運(yùn)維的技術(shù)。Facebook實(shí)現(xiàn)了蒸發(fā)冷卻，它從外部吸入空氣，過濾后，向空氣中噴灑純凈水霧，通過水蒸發(fā)來實(shí)現(xiàn)制冷。冷卻后的空氣經(jīng)過另一系列的過濾器，以確?？諝膺M(jìn)入數(shù)據(jù)中心前達(dá)到規(guī)定的溫度和濕度。如果在允許的溫度范圍內(nèi)，可以使用外部空氣過濾系統(tǒng)，而外部空氣過熱時(shí)則可以關(guān)閉系統(tǒng);如果外部空氣太冷，可以用數(shù)據(jù)中心的空氣來混合。最初，根據(jù)ASHRAE的建議，空氣被冷卻到80.6℉（譯者注：27℃），F(xiàn)acebook準(zhǔn)備根據(jù)視頻錄制時(shí)的實(shí)施結(jié)果將氣溫調(diào)整到85℉（譯者注：29.4℃）。此外，F(xiàn)acebook數(shù)據(jù)中心采用LED照明、雨水沖廁和其他同類措施進(jìn)一步降低數(shù)據(jù)中心的整體能耗。

Intel performed a proof of concept experiment with even a wider range of temperatures than ASHRAE recommends, on 900 production servers in a high density/high utilization environment. Half of the servers were maintained as usual with chilling units as a control, the other half used free air cooling. The half using free air cooling experienced a wide variation of temperatures (64 – 92 degrees F) and humidity levels (4-90% relative humidity) and the entire data center had a layer of dust settle over the equipment. In addition to a 74% reduction in energy consumption, the free air cooling side experienced no significant increase in server failure rates despite the widely variable environment that they were housed in [17].

英特爾在一個(gè)高密度/高利用率環(huán)境中的900臺生產(chǎn)服務(wù)器上進(jìn)行了一個(gè)概念驗(yàn)證實(shí)驗(yàn)，其溫度范圍甚至比ASHRAE建議的范圍更廣。一半的服務(wù)器像往常一樣使用冷水機(jī)組作為控制，另一半使用自然空氣冷卻。使用自然空氣冷卻的那一半在溫度(64-92℉，譯者注：17.8℃-33.3℃)和濕度(4-90%相對濕度)范圍內(nèi)變化很大，整個(gè)數(shù)據(jù)中心設(shè)備上覆蓋了一層灰塵。盡管服務(wù)器所處的環(huán)境變化很大，但是除了減少了74%的能源消耗之外，自然空氣冷卻的那部分服務(wù)器的故障率沒有顯著增加。

Using CFD analysis of free air cooling, a case study by Gebrehiwat et al [27] shows that when strong server fans are in use, it may not be necessary to use blowers in the power and cooling modules [27]. Conversely, Facebook has demonstrated the ability to reduce the use of the server fans by pressurizing the data center and increasing the wind power via large energy efficient fans, which distribute the cooler air [19]. Either way shows promise and appear to suggest that using both at full scale may be an unnecessary energy drain.

Gebrehiwat等人通過對一個(gè)自然空氣冷卻案例的CFD分析表明，在使用強(qiáng)大的服務(wù)器風(fēng)扇時(shí)，可能不需要在電源和制冷模塊中使用風(fēng)機(jī)。相反，F(xiàn)acebook通過高能效風(fēng)扇增加風(fēng)力以及給數(shù)據(jù)中心加壓，優(yōu)化冷空氣分配，展示了減少服務(wù)器風(fēng)扇使用的能力。無論采用哪種方式，似乎都表明完全使用這兩種方法可能是一種不必要的能源消耗。

COMPREHENSIVE TOOLS

綜合工具    

These tools do not neatly categorize into one of the previous topics, since they optimize multiple areas. 

由于這些工具實(shí)現(xiàn)了多個(gè)不同領(lǐng)域的優(yōu)化，所以難以將他們清晰地分類到前面的某一個(gè)主題中。

CoolEmAll [1] enables a comprehensive analysis of data center efficiency by providing a Simulation, Visualization, and Decision support toolkit (SVD Toolkit) integrating models of: application/server workload scheduling and power requirements, equipment characteristics, and cooling. CoolEmAll will enable data center designers and operators to analyze, simulate (via CFD), and optimize efficiency of existing and planned data centers. The project will define data center building blocks that design hardware specifications, geometrical models, and then project energy efficiency metrics. It will also deliver a toolkit that can be used to analyze andoptimize data centers that are built with these building blocks. CoolEmAll distinguishes their tool by optimizing low and variable loads in addition to peak loads. Similar to the Open Compute project, this one will also design pre-configured and ready to use components for data centers.

CoolEmAll軟件通過提供一個(gè)仿真、可視化和決策支持工具包(SVD工具包)以全面分析數(shù)據(jù)中心的效率，該工具包集成了以下模型:應(yīng)用程序/服務(wù)器工作負(fù)載調(diào)度和電源需求、設(shè)備參數(shù)和制冷。CoolEmAll使數(shù)據(jù)中心設(shè)計(jì)運(yùn)維人員能夠通過CFD分析、模擬現(xiàn)有和計(jì)劃中的數(shù)據(jù)中心，并優(yōu)化其效率。該項(xiàng)目將定義數(shù)據(jù)中心建筑模塊的設(shè)計(jì)硬件規(guī)格、幾何模型和項(xiàng)目能效指標(biāo)。項(xiàng)目還提供了一套工具包，通過建立建筑模塊來分析和優(yōu)化數(shù)據(jù)中心。除了峰值負(fù)載，CoolEmAll工具還可分別用于優(yōu)化低負(fù)載和可變負(fù)載。與開放計(jì)算項(xiàng)目類似，該軟件的組件還可以用于設(shè)計(jì)已有的或預(yù)制數(shù)據(jù)中心。

The EoD Designer [8] software also acts as both a planning and optimization tool for data centers allowing the user to graphically design different data center configurations. It is coupled with a mathematical model to compute the total energy consumed by the data center. This mathematical model is based on six components of the data center: voltage transformer, uninterruptable power supply, IT equipment, computer room air handler, chiller and cooling tower. It then integrates this mathematical model into this software tool, which can simulate different data center configurations and lead to a minimal total energy consumption within the data center. The tool ships with several default data center components, but also allows the operator to define custom components. Detailed reports can be generated and display data based on monthly or yearly time intervals.

EoD Designer軟件同樣可以作為數(shù)據(jù)中心的規(guī)劃和優(yōu)化工具，允許用戶對不同配置的數(shù)據(jù)中心開展圖形化設(shè)計(jì)。它與數(shù)學(xué)模型相結(jié)合來計(jì)算數(shù)據(jù)中心消耗的總能量。該數(shù)學(xué)模型基于數(shù)據(jù)中心的六個(gè)組成部分：變壓器，UPS，IT設(shè)備，機(jī)房空調(diào)，冷水機(jī)組和冷卻塔。然后將該數(shù)學(xué)模型集成到該軟件工具中，可以模擬不同的數(shù)據(jù)中心配置，從而將數(shù)據(jù)中心總能耗降至最低。該工具附帶幾個(gè)默認(rèn)的數(shù)據(jù)中心組件，但也允許用戶自定義組件。該軟件可以根據(jù)每月或每年的時(shí)間間隔生成詳細(xì)的報(bào)告并顯示數(shù)據(jù)。

GDCSim [9] identifies a full set of features that would represent a holistic data center simulator including: automated processing, online analysis, iterative design, thermal analysis, workload management and cyber-physical interdependency. GDCSim proposes a simulation tool that simulates the physical behavior of the data center under different management techniques. The tool will have three components: CFD simulator, resource management, and a simulator. In operation, the three components work together in an iterative fashion to design the data center and determine the ideal resource management plan. This functionality was demonstrated in two case studies, which continuously monitored the Energy Reuse Effectiveness (ERE) during the design iterations. Both case studies demonstrated that the tool had a positive influence on the ERE measured in the data center [9].

GDCSim軟件標(biāo)識了一組完整的特性，這些特性表示一個(gè)完整的數(shù)據(jù)中心模擬器，包括：自動處理、在線分析、迭代設(shè)計(jì)、熱分析、負(fù)載管理、網(wǎng)絡(luò)物理系統(tǒng)依存。GDCSim提出了一種仿真工具，可以模擬數(shù)據(jù)中心在不同管理技術(shù)下的物理行為。該工具將包括三個(gè)組件:CFD仿真器、資源管理和模擬器。在運(yùn)行中，三個(gè)組件以相互迭代的方式工作，設(shè)計(jì)數(shù)據(jù)中心并確定理想的資源管理計(jì)劃。此功能在兩個(gè)研究案例中得到了演示，它們在設(shè)計(jì)迭代期過程中持續(xù)監(jiān)視再生能源利用效率(ERE)。兩個(gè)案例研究都表明，該工具對數(shù)據(jù)中心測量的ERE值具有積極影響。

Goiri et al [5] developed a financial analysis tool to model power consumption for a virtualized data center. This tool optimizes resource management via scheduling, cost analysis and fault tolerance implementations on two experimental virtualized data centers. The scheduling algorithm they developed assesses each decision based on maximizing the data center provider’s profit. This algorithm uses the cost associated with power consumption as a variable in maximizing the provider’s profit, which correlates to reducing power consumption. The scheduling policy determines the host allocation for each virtualized machine, ensuring fault tolerance and determining an on/off cycle for each node. The researchers then compared their scheduling policy against commonly used ones and showed a 15% benefit to the provider’s profits when using their policy.

Goiri等人開發(fā)了一個(gè)財(cái)務(wù)分析工具為虛擬數(shù)據(jù)中心功耗建模。該工具通過調(diào)度、成本分析、在兩個(gè)實(shí)驗(yàn)性虛擬化數(shù)據(jù)中心實(shí)現(xiàn)容錯(cuò)來優(yōu)化資源管理。他們開發(fā)的調(diào)度算法基于最大化數(shù)據(jù)中心提供商的利潤來評估每個(gè)決策。該算法將與功耗相關(guān)的成本作為最大化供應(yīng)商的利潤的一個(gè)變量，這與降低功耗有關(guān)。調(diào)度策略確定每個(gè)虛擬機(jī)的主機(jī)分配，確保容錯(cuò)，并確定每個(gè)節(jié)點(diǎn)的啟/停周期。然后，研究人員將他們的調(diào)度策略與常用的調(diào)度策略進(jìn)行了比較，發(fā)現(xiàn)當(dāng)使用他們的策略時(shí)，供應(yīng)商可獲得15%的收益利潤。

ANALYSIS

分析

There appears to be a gap in tools for businesses with small scale data centers [21]. These data centers do not have the expertise nor money to pay for expensive CFD tools, nor do they have the resources for a complete redesign, or to implement costly, large scale equipment. IT personnel often lack time and expertise in industrial engineering required to perform particularly the thermal analysis aspects.

對于擁有小型數(shù)據(jù)中心的企業(yè)來說，似乎在工具方面存在差距。這些數(shù)據(jù)中心沒有專業(yè)知識，沒有資金購買昂貴的CFD工具，也沒有資源來進(jìn)行徹底的重新設(shè)計(jì)，或完成昂貴的大規(guī)模設(shè)備。IT人員通常缺乏執(zhí)行工業(yè)工程熱分析方面的所需的時(shí)間和專業(yè)知識。

One option is to consolidate small data centers into larger ones. This may present better options for optimization, such as through using hosted data centers and cloud computing. Massive data centers can optimize for best practices, such as those used by Facebook, Google and Intel.

選擇之一是將小型數(shù)據(jù)中心合并為大型數(shù)據(jù)中心。通過使用托管數(shù)據(jù)中心和云計(jì)算，可能是更好的優(yōu)化選項(xiàng)。大型數(shù)據(jù)中心可以像Facebook、谷歌和英特爾那樣優(yōu)化最佳實(shí)踐。

However, not all businesses are prepared to move to this type of data center solution. For these businesses, a gap exists which is not solved by many of the above-mentioned tools, since CFD tools are expensive, complex to run, and protected by patents. Low cost improvements in airflow can include monitoring the temperature throughout the data center by placing thermometers in key locations and by using sheet metal or meat locker curtains to contain or control warm air flow. The analysis could calculate the potential savings of in- rack based air conditioning or free air cooling, and indicate permissible temperature ranges for various equipment types.

然而，并不是所有企業(yè)都準(zhǔn)備遷移到這種類型的數(shù)據(jù)中心解決方案。對于這些企業(yè)來說，由于CFD工具價(jià)格昂貴、運(yùn)行復(fù)雜、受專利保護(hù)，上述許多工具都無法彌補(bǔ)這一差距。在氣流方面的低成本改進(jìn)可以包括通過在關(guān)鍵位置放置溫度計(jì)來監(jiān)視整個(gè)數(shù)據(jù)中心的溫度，并通過使用金屬板或冷庫隔簾來容納或控制熱氣流流動。該分析可以計(jì)算基于空調(diào)或自然空氣冷卻的機(jī)架的可能節(jié)約情況，并明確各種設(shè)備類型的允許溫度范圍。

A database tool available for smaller scale data centers could include efficiency metrics to compare with best in class, and modeling virtual machines. In addition, small to mid-size data centers should have the ability to inventory, analyze and rationalize their application portfolio. Consolidation could occur when redundant applications and resources are found. Infrastructure upgrades and application consolidation could be analyzed for energy efficiency to justify infrastructure changes. A third feature would compare the energy consumption of existing infrastructure equipment versus newer energy star equipment and resulting costs.

對于較小規(guī)模的數(shù)據(jù)中心可用的數(shù)據(jù)庫工具能提供一流的效率指標(biāo)和虛擬機(jī)建模。此外，中小型數(shù)據(jù)中心應(yīng)該具有編目、分析和合理化應(yīng)用組合程序的能力。冗余應(yīng)用程序和資源可以整合?？梢詫A(chǔ)設(shè)施升級和應(yīng)用程序整合進(jìn)行分析，以提高能源效率，從而證明基礎(chǔ)設(shè)施的改變是合理的。第三個(gè)特點(diǎn)是比較現(xiàn)有基礎(chǔ)設(shè)施設(shè)備與新能源之星設(shè)備的能耗及最終成本。

CONCLUSION

結(jié)論

The demand for data center storage is expected to continue to rise and thus the topic of data center efficiency will continue to be an important one. Google has set a high standard for how to approach data center efficiency. We have explored methods of efficiency and considered how they could be adapted for smaller data centers.

對數(shù)據(jù)中心存儲的需求預(yù)計(jì)將持續(xù)上升，因此數(shù)據(jù)中心效率問題將依然是一個(gè)重要的課題。對于如何提高數(shù)據(jù)中心的效率，谷歌設(shè)定了很高的標(biāo)準(zhǔn)。我們已經(jīng)探索了一些提高效率的方法，并思考如何將其適用于更小的數(shù)據(jù)中心。

翻譯：

何海

DKV（Deep Knowledge Volunteer)計(jì)劃精英成員

中國空氣動力研究與發(fā)展中心，工程師

編輯：

梁鴻雁

中能測（北京）科技發(fā)展有限公司秘書處處長