数据中心主要发展趋势（可持续数据中心的兴起）
数据中心主要发展趋势（可持续数据中心的兴起）Sustainability is now a top-of-mind issue in data center construction and management as cloud computing and digital transformation fuel the world’s growing thirst for processing power. Energy Innovation Policy & Technology reports that global IP traffic increased ten-fold between 2010 and 2018 the number of compute instances running on the world’s servers increased more than six-fold and Data Ce
可持续数据中心的兴起
The Rise of the Sustainable Data Center
By Paul Gillin
译 者 说
智能电网充分利用园区内UPS的电池容量&简化供电路径降低用水量是未来运营数据中心最值得创新点之一。
可持续数据中心的兴起
The Rise of the Sustainable Data Center
由于云计算和数字化转型刺激了世界持续增长的能源消耗的渴望，可持续发展成为当今数据中心建设和管理的首要问题。能源创新政策及技术有限责任公司(Energy Innovation Policy & Technology)报告：2010年到2018年之间，全球IP流量增长了10倍，运行在服务器上的算例增长了6倍多，而数据中心的存储能力暴涨了25倍。如今 某些世界上最大的数据中心所消耗的电能足以为8万个家庭供电。
Sustainability is now a top-of-mind issue in data center construction and management as cloud computing and digital transformation fuel the world’s growing thirst for processing power. Energy Innovation Policy & Technology reports that global IP traffic increased ten-fold between 2010 and 2018 the number of compute instances running on the world’s servers increased more than six-fold and Data Center storage capacity skyrocketed by a factor of 25. Some of the world’s largest data centers now consume enough electricity to power 80 000 homes.
而且，计算负荷只会越来越大。海量的GPU和CPU被部署用以支撑数量迅速增长的人工智能应用。配备5G无线网络的智能设备将催生小型但数量众多的边缘数据中心。
And workloads are only going to grow more. A profusion of GPUs and CPUs are being deployed to support a burgeoning number of artificial intelligence applications. Smart devices equipped with 5G wireless networks will give rise to smaller but numerous edge data centers.
互联设备数量的增长不仅推动了运行它们的电力需求，而且推动了连接它们的数据中心的需求。维谛估计，到2026年5G的使用可能会使整个网络的能耗增加150%到170%，其中最大的增长来自于大型、节点型和网络型数据中心领域。一项2017年的研究预测：到2025年 通讯业使用的电能会占到全球电能总量的20%。同时，产生的排放到2040年将占全球的14%，和今天美国的排放量相当。
The growth of connected devices is driving not only demand for power to operate them but also data centers that connect to them. Vertiv estimates that the move to 5G is likely to increase total network energy consumption by between 150% and 170% by 2026 with the largest increases coming in macro node and network data center areas. One 2017 study forecast that the communications industry could use 20% of the world’s electricity by 2025 and generate 14% of global emissions by 2040 or about as much as the entire U.S. does today.
所有这些都是在这样的背景下发生的，全球对于保证水供应安全以及限制化石能源使用的关注日益增加。某种程度看，这两者之间存在着平衡。数据中心所消耗的电力约40%是用于冷却设备的空调。通过蒸发冷却技术的使用，可以极大程度的降低这一数字， 但是其收益是以增加用水量及保持水供应的清洁所需的额外费用和工艺步骤为代价的。要实现最佳的可持续发展方案需要平衡好这两个变量。
All of this is occurring against the backdrop of growing global concerns about keeping water supplies safe and limiting the use of fossil fuels. To some degree there is a trade-off between the two. About 40% of the power consumed by data centers goes to the airconditioning used to cool equipment. That figure can be significantly reduced through the use of evaporative cooling technologies but the gains come at the expense of increased water usage and additional costs and steps needed to keep water supplies clean. Achieving the most sustainable solution requires balancing both variables.
可持续发展的新思路
New Ideas About Sustainability
超大型数据中心运营商已经广泛接纳了可持续运营的目标，并且在贡献了一些重要创新的同时，显著降低了能耗。例如，如今谷歌的数据中心相比于5年前算力平均增加了7倍多，但能耗并未增加。据美国数据中心用能报告，在2010到2014年间，数据中心的能耗仅增加了4%，远比5年前的24%低得多。预计未来能源使用的增长速度会更慢。
Hyperscale data center operators have broadly embraced the goal of sustainable operations and have contributed some important innovations while significantly reducing their own power use. For example Google’s data centers today support an average of seven times more computing power than they did five years ago but use no more electricity. data center electricity consumption increased just 4% between 2010 and 2014 far below the 24% increase of the previous five years according to the U.S. Data Center Energy Usage Report. Energy use is expected to increase at an even slower rate in the future.
这种增长放缓的部分原因是服务器虚拟化的广泛使用，减少了数据中心所需的服务器数量的同时提高了服务器的利用率。利用与可再生能源供应商之间的电力购买协议，创造一种具有成本效益的能源替代方案，超大型数据中心在这方面也一直处于领导地位，建立自己的可再生能源基础设施的同时，为可持续能源投资提供资金。像谷歌的“碳智能”数据中心这样的创新设计，可以自动转移工作负载，以最大限度地利用零碳电能。
Part of this slowing growth is due to the widespread use of server virtualization which has reduced the number of servers needed in data centers while driving up utilization rates. Hyperscalers have also been leaders in the use of power purchase agreements with renewable energy providers to create a cost-effective alternative to building out their own renewable infrastructure while funding investments in sustainable sources. Innovations like Google’s “carbon-intelligent” data center design shifts workloads automatically to maximize the use of carbon-neutral power sources.
尽管数据中心行业对电力有着巨大的需求，但大多数数据中心运营商对大规模使用可替代能源持谨慎态度。这并不意味着他们选择沉默，而是需要在可持续性和弹性之间进行平衡。
But despite the industry’s voracious appetite for power data center operators for the most part have moved cautiously to adopt alternative energy sources on a large scale. This does not reflect reticence their part but rather the need to balance sustainability with resiliency.
数据中心需要高质量且可预测的电能。低至30毫秒的电压波动和跌落能够损坏IT设备，从而引发代价高昂的宕机。Statista报告称，2019年企业服务器宕机成本的中位数在40万至50万美元之间，其中15%的企业报告宕机成本高于500万美元。对于多租户数据中心的运营商而言，潜在宕机成本要高得多，包括业务损失、未能满足服务水平协议的罚款及法律风险。
Data centers require power that is both high-quality and predictable. Voltage fluctuations and dropouts of as little as 30 milliseconds can damage IT equipment triggering expensive outages. Statista reported that the median cost of enterprise server downtime in 2019 was between $400 000 and $500 000 with 15% of enterprises reporting costs of greater than $5 million. For operators of multi-tenant data centers the potential costs are much higher including lost business penalties for failure to meet service level agreements and legal exposure.
像太阳能和风能等可持续性能源本质上是不可预测的，因为它们依赖于天气条件。燃料电池显示出前景，但它们目前价格昂贵且不成熟。太阳能和风能可用作辅助电源，但存储足够电力以避免突发事件，加上选址、安装和设备维护的成本太高，以至于对许多运营商来说是不切实际的。电力购买协议可以提供许多相同的好处，而无需高昂的投资成本。
Sustainable energy sources such as solar and wind are inherently unpredictable because of their dependence upon weather conditions. Fuel cells show promise but they are currently expensive and immature. Solar and wind can be useful as an ancillary power supply but the cost of storing enough power to insure against disruptions—combined with the cost of siting installing and maintaining equipment—is so high as to be impractical for many operators. Power purchase agreements can provide many of the same benefits without the high capital costs.
对清洁而稳定的能源需求是大多数数据中心使用多个不间断电源的原因之一。UPS不仅可以防止断电，而且可以消除直接来自电网的电力波动。不幸的是，大多数数据中心UPS在设计上并未得到充分利用。这些设备以约80%的容量运行的时候效率最高，但数据中心运营商通常以不超过20%的容量运行它们，以确保在发生断电时的可用性。这就导致了大量的电池容量被堂而皇之地浪费了。
The need for stable and clean power is one of the reasons most data centers maintain multiple uninterruptible power supplies. UPSes not only protect against outages but eliminate the fluctuations of power taken directly from the grid. Unfortunately most data center UPSes are under-utilized by design. The units are most efficient when operating at about 80% capacity but data center operators typically run them at no more than 20% capacity to ensure availability in the event of an outage. That adds up to a substantial amount of battery capacity that is effectively wasted.
正如Iron Mountain的环境、社会和治理战略副总裁，Kevin Hagen，告诉Data Center Knowledge，“UPS电池就是无用的资产，我们花费数百万美元的保单只是放在那希望我们不要使用它。那是一笔可怕的开销。”
As Kevin Hagen Iron Mountain’s vice president of environment social and governance strategy told Data Center Knowledge UPSes batteries are “useless capital. We’re spending millions of dollars of insurance policy just sitting there in the hope we don’t use it. That’s a terrible use of money.”
这种未使用的容量可能会成为稳定电网、促进可持续能源使用并为数据中心运营商和他们购买电力的市政公司创造经济利益的资源。未使用的UPS电池容量可用于使可持续性能源更加实用，甚至改善电网的整体健康状况。
This unused capacity could potentially be a resource to stabilize the electrical grid boost usage of sustainable energy and create financial benefits for both data center operators and the utilities from which they buy their power. Unused UPSes battery capacity could be used to make sustainable energy sources more practical and even improve the overall health of the energy grid.
更智能的电网
A Smarter Grid
为美国提供大部分电力的电网是在19世纪90年代设计的，并在此基础上不断改进，但受其老旧设计的限制。在电网诞生的那个年代，家庭用电需求并不大。没有人设想过500瓦的个人电脑、电烤箱和电池驱动的汽车，更不用说100兆瓦的数据中心了。
The grid that provides most of the electrical power in the U.S. was designed in the 1890s and has been incrementally improved upon but it is limited by its aging design. power needs in the home were modest in the days the grid was designed. No one conceived of 500-watt PCs electric ovens and battery-powered cars much less 100-MEGAWATT data centers.
电网被设计为从电力公司到按月计费的用户之间的单向路径。通过分区，故障影响范围已稳步缩小，但今天的停电依然可能影响很多人。或许，今天电网最大的短板就是它的“单向性”。它被设计为电力公司与其用户之间的管道而已。虽然在过去的时间里，电网已经升级成允许双向传输，但用户之间仍然难以相互之间交易电力。而这越来越成为一种可能，因为许多家庭和企业已经变成了微型发电站，通常生产的电力超过了他们自身的需求。
The grid was designed to be a one-way path from the utility to the customer who was billed monthly. The scope of power outages has been steadily reduced through sub-zoning but failures today still have the potential to affect a lot of people. Perhaps the greatest shortcoming of the current power grid is its “one-wayness”. It was designed as a conduit between a utility and its customers. While the network has been upgraded over time to permit two-way transmission there is still no easy way for customers to exchange electricity with each other. This has increasingly become a liability as many homes and businesses have become miniature power stations often generating more electricity than they need.
美国能源局的智能电网计划以类似于互联网的方式重新构想了电网。当前电网上的9200套发电机组将辅以数千个变电站，这些变电站可以根据需要在节点之间智能地分配电力。客户可以组成一个群体来共享他们产生的电力，以降低成本或提高可靠性。停电将持续几分钟而不再是几小时了。电网的整体弹性将得到改善，可持续性能源变得更加实用。
The US Department of Energy’s Smart Grid initiative reimagines the power grid in a manner similar to the Internet. The 9 200 electric generating units on the current grid would be complemented by thousands of switching stations that intelligently route power between nodes as needed. Customers could form collectives to share the power they generate to lower costs or improve reliability. Blackouts would last minutes instead of hours. The overall resilience of the grid would improve and sustainable sources become more practical to use.
领先的数据中心技术公司的研究人员正在开发可以使UPS成为电网上智能节点的技术。平均80%未使用的数据中心UPS可能成为网络上的一个节点，提供临时性电力存储和供应。UPS电池不仅可以在停电时用作备用电源，而且由于UPS固有地提供更稳定电力的能力，它们可以帮助补偿整个电网的功率和电量的波动。
Researchers at leading data center technology companies are developing technologies that can make UPSes smart nodes on the energy grid. The 80% of the average data center UPSes that go unused could become a node on the network offering temporary power storage and provisioning. UPS batteries could not only serve as backup power sources in case of an outage but because UPSes inherently deliver a more stable flow of power they could help to compensate for fluctuations in power levels and quality across an entire grid.
例如，为一个15兆瓦的数据中心供电5分钟所需的UPS电池容量可以为大约15000个家庭供电相同的时间。虽然五分钟看起来不是很长，但对于电力公司来说，在高需求或中断期间重新平衡供电能力已经足够了。通过智能电网，电力可以很快地重新分配，大多数消费者甚至看不到自己的灯在闪烁。UPS还可以减轻电涌对精密电子设备的损害，提高电网的整体质量。
For example the UPS battery capacity needed to power a 15-MEGAWATT data center for five minutes could supply electricity to approximately 15 000 homes for the same duration. While five minutes doesn’t seem like a long time it is more than enough for utilities to rebalance capacity during periods of high demand or disruption. Power could be rerouted through the smart grid so quickly that most consumers wouldn’t even see their lights flicker. UPSes can also mitigate the impact of power surges that damage delicate electronic equipment enhancing the overall quality of the grid.
目前参与智能电网实验的数据中心运营商屈指可数，他们大部分在欧洲。然而，随着技术的进步，再加上成本的降低，这一选择将在几年内得到广泛应用。引领这一潮流的数据中心技术公司将首先为他们的客户带去这些好处。一些公司甚至正在开发赋能现有的UPS成为智能电网智能集线器的改造技术。
There are currently only a small number of data center operators engaged in smart grid experiments most of them in Europe. However improvements in technology combined with lower costs will make this option widely available within a few years. The data center technology companies that are leading the charge will bring these benefits to their customers first. Some are even developing technologies that will enable existing UPSes to be retrofitted as smart hubs on the smart grid.
应对耗水量问题
Tackling Water Usage
2019年初结束的长达7年的加州干旱和随后发生的野火只是最近发生的两起事件，让人们关注到缺水加剧的深远后果。气候变化对水源供应影响的不确定性，加剧了此类担忧。即使没有环境变化，世界水资源消耗的速度也越来越快。全球人口在过去40年翻了一番，但用水量却翻了两番。水资源组织预测，到2030年，全球水需求可能会超过可持续消耗量的40%。
The seven-year-long California drought that ended in early 2019 and the wildfires that ensued are just two recent events that have cast a spotlight on the far-reaching consequences of worsening water shortages. Those concerns have been compounded by lack of certainty about what the impact of climate change will be on water supplies. Even without environmental change the world is consuming water at ever-faster rates. Global population has doubled over the past 40 years but use of water has quadrupled. The Water Resources Group forecasts that global water demand may outstrip sustainable use by 40 percent as soon as 2030.
数据中心受到了特别的关注。2020年，仅美国，就将消耗大约1740亿加仑的水。而一个15兆瓦的数据中心每天需要消耗多达36万加仑的水。担心公共资源缺乏的州和地方机构正在施加监管压力，而像绿色和平(Greenpeace)这样的激进组织则特别强调，需要关注数据中心对电力和水的使用。所有这一切发生的同时，由于制冷剂的使用限制，冷却方式的选择却越来越少。
Data centers are under particular scrutiny. In the U.S. alone they are expected to consume an estimated 174 billion gallons of water in 2020. A 15-megawatt data center can use up to 360 000 gallons of water a day. Regulatory pressure is being turned up by state and local bodies that are concerned about scarcity of public resources while activist groups like Greenpeace have put a special emphasis on drawing attention to data centers’ use of both power and water. All this is happening as cooling options are narrowing due to restrictions on the use of refrigerants.
除了数据中心自身的整体增长，冷却需求也受到了机器学习、加密货币挖掘和物联网设备等新应用的驱动。人工智能计算负荷大量使用耗电大的图形处理单元，而人工智能的训练算法需要多达数天的繁重计算过程。
In addition to overall data center growth demand for cooling is being driven by new applications like machine learning cryptocurrency mining and internet of things devices. Artificial intelligence workloads make extensive use of power-hungry graphics processing units and AI training algorithms can require days of heavy processing.
数据中心的用水有广泛影响。它影响着当地供水的质量和可用性，特别是涉及到地下水时。这会招致更多的监管审查 运营商可能不得不进行昂贵的基础设施投资来清洁循环水。
Data center water use has broad impacts. It affects the quality and availability of local water supplies particularly when groundwater is involved. It invites increased regulatory scrutiny and operators may have to make expensive infrastructure investments to clean recycled water.
超大型运营商还引领创新的冷却策略，率先使用自然风冷，提高服务器运行的入口空气温度，并试验新技术，如间接蒸发制冷、制冷剂节能、芯片级液冷、浸没式液冷和背板水冷单元。然而，在满足日益增长的由高密计算基础设施带来的制冷需求的同时，没有明确的解决方案来减少水的使用。
Hyperscale operators are also leading the charge toward innovative cooling strategies pioneering the use of free-air cooling raising operating temperatures for server inlet air and experimenting with new technologies like indirect air cooling refrigerant economization water to the chip immersion cooling and rear-door chilling units. However there is no clear solution to reducing water usage while meeting the growing cooling requirements of ever-denser compute infrastructure.
而且这个等价关系比水冷还是风冷的权衡要复杂很多。事实上，风冷实际会增加数据中心整体用水量。空调是个电力消耗大户 根据The Engineering Mindset的说法，一个典型的数据中心有35%的电力用于暖通空调。
And the equation is more complicated than just the tradeoff of water versus air cooling. In fact air cooling can actually increase overall water consumption. Air conditioning is an enormous consumer of power. A typical data center uses 35% of its power for HVAC according to The Engineering Mindset.
水在电力消耗中的作用经常被忽视。很少有人知道，在美国和欧洲，发电消耗了一半以上的水。用煤或核燃料发电一千瓦需要大约15加仑的水。这意味着对于一些数据中心来说，减少用水最有效的方法就是减少电力消耗。平衡在于寻求有效的技术，减少甚至消除对水的消耗。
The role of water in power consumption is often overlooked. Few people are aware that power generation consumes more than half the water in the U.S. and Europe. Generating a single kilowatt of electricity with coal or nuclear fuel requires about 15 gallons of water. That means that the most effective way for some data centers to reduce water usage is to cut back on power consumption. The balance comes in seeking efficient technologies that reduce or even eliminate the need for water consumption.
寻求平衡
Seeking a Balance
最好的策略是寻求平衡。从2006年开始，绿色网格的PUE评级已经被用作衡量数据中心效率的黄金准则。PUE的计算方法是将进入设备的总电能除以IT设备使用的电能，目标是尽可能达到1:1。尽管绿色网格已声明PUE严格来说是数据中心运营商跟踪其自身能源效率的指标，但它已被美国国家标准学会/美国暖通学会/美国照明工程学会ANSI/ASHRAE/IES标准90.1-2013采用，从而激励行业使用水进行冷却。通过将负荷从电力转移到水，数据中心可以提高他们的PUE，从而更好地符合规定，因为水的消耗计算起来不可见。
The best strategy is to seek a balance. Since 2006 Green Grid’s Power Usage Efficiency (PUE) rating has been used as the gold standard for data center efficiency. PUE is calculated by dividing the amount of power entering a facility by the amount used to support the IT load with the goal being to get as close to a 1:1 ratio as possible. Although the Green Grid has stated that PUE was strictly meant to be a metric for data center operators to track their own power efficiency it was adopted in the ANSI/ ASHRAE/IES Standard 90.1-2013 thereby providing an incentive for the use of water cooling. By shifting the burden from electricity to water facilities can improve their PUE and thereby better comply with regulations because water consumption is invisible to the calculation.
2011年，作为与PUE相对应的一个必然结果，绿色网格引入了用水效率(WUE)指标。用水效率的计算方法是将现场用于冷却、调节湿度和发电的用水量除以IT设备的能源使用量。这是为了让数据中心的管理人员了解用水量对当地电网的影响。
In 2011 the Green Grid introduced the Water Usage Effectiveness (WUE) metric as a corollary to PUE. WUE is calculated by dividing the amount of water a site uses for cooling regulating humidity and producing electricity by IT equipment energy usage. It’s meant to enable data center managers to understand the effect water consumption has on the local electric grid.
数据中心运营商需要同时考虑这两个指标，以实现对他们的营运底线和可持续发展目标都有利的平衡。投资像风能和太阳能这样的可替代能源，这两种能源生产的用水量都比煤电厂或核电厂少得多，可以帮助降低这两个指标的分值。
Data center operators need to consider both metrics in achieving a balance that is good for both their bottom lines and sustainability objectives. Investments in alternative energy sources like wind and solar both of which use far less water than coal-or nuclear-powered plants can help reduce both scores.
最好的做法是将热管理技术的选择映射到实际案例中。最少电力消耗的低水或零水的解决方案是最理想的。已有几种冷却的可替代方案既能节水又能节电。
The best course of action is to map the choice of thermal technology to the use case. Low- or zerowater solutions that also minimize power use are the most desirable. There are several types of cooling alternatives that are both water- and power-efficient.
直接蒸发制冷
Direct evaporative cooling
直接蒸发制冷利用水的直接蒸发以低能耗产生显著的冷却和加湿效果。使用像CELdek或GLASdek这样的亲水性介质作为基材，让大量空气接触蒸发的水。这被广泛认为是空气冷却和加湿的最简单、最经济的方法。这种技术是由那些愿意让数据中心内部的空气跟随室外空气的扰动，并愿意冒险从室外直接引入空气的运营商开发的。为了将风险降到最低，直接蒸发系统必须包含完整而精简的备份系统，以防室外空气不适合数据中心，就像加州和美国其他地区发生过的火灾一样。
Direct evaporative cooling uses direct evaporation of water to produce significant cooling and humidification with low energy consumption. A wetted media like CELdek or GLASdek is used as a substrate to allow a large volume of air contact evaporating water. This is widely regarded as the simplest most cost-effective method of cooling and humidification for air. This technology is deployed by those operators that are willing to let the air in the data center fluctuate with the outdoor air and are willing to take the risk of bringing in air from outside. To minimize these risks direct evaporative systems must include full compressorized backup systems in the event outdoor air is not suitable for the data centers as is the case with fires that have occurred in California and other parts of the U.S.
间接蒸发制冷
Indirect evaporative cooling
间接蒸发制冷利用接触聚合物热交换器不同侧面的两个方向相反的气流制冷。交换器外壁接触的是在送入到作用空间之前需要调节温度的空气，而内壁与来自周围环境或建筑物排出的空气接触。当喷洒在热交换器内壁上的水蒸发时发生冷却，从而将冷却效果传递至热交换器的外壁。使得与热交换器外壁接触的气流被冷却。这项技术尽管高效，但是和直接蒸发制冷相比，由于室内侧和室外侧之间的热交换器的存在，间接蒸发制冷会消耗更多的水。它的好处是消除了将室外空气直接引入到数据中心内部相应的风险，以及按照数据中心满负荷的需要，按比例备份系统或备份机械制冷能力的需要。
Indirect evaporative cooling uses two opposing airstreams that contact a different side of a polymer heat exchanger. The outer wall of the exchanger contacts air that needs to be conditioned before it is delivered to the occupied space. The inner wall is in contact with air that comes from the ambient environment or building exhaust. Cooling occurs when water sprayed on the interior wall of a heat exchanger evaporates imparting a cooling effect to the outer wall of the heat exchanger. This allows the airstream that contacts the outer wall to be cooled. This technology while efficient can consume more water than direct evaporative cooling due to the heat exchanger being between the indoor and outdoor air. The benefit of this technology over direct evaporative cooling is that it eliminates the risks associated with bringing outdoor air into the data center and the need to size backup or compressorized cooling to the full load of the data center.
蒸发式自然冷却冷水机组
Evaporative freecooling chillers
蒸发式自然冷却冷水机组使用经过湿垫或雾化喷雾系统的环境空气，以延长冷水机组可以在自然冷却模式下运行的时间。在温度较高的日子里，蒸发冷却的空气进入排热盘管(冷凝器)，从而降低能源成本。采用绝热自然冷却或蒸发自然冷相结合的混合运行模式，同时降低冷凝空气温度，不仅提供了更高效的冷水机组的运行条件，还降低了冷水机组运行的峰值功率要求，同时降低了支持冷却负荷所需的发电和配电需求。
Evaporative freecooling chillers use ambient air that is passed through wet pads or atomizing spray systems to extend the hours the chiller can operate in free cooling mode. During warmer days evaporative cooled air is injected into the heat rejection (condenser) coils thereby reducing energy costs. The combination of operating modes between freecooling with the adiabatic or evaporative system and lowered condensing air temperature not only provides for a more efficient chiller operation but also reduces the peak power requirements for the chiller to operate reducing generator and power distribution needed to support the cooling load.
制冷剂节热器
Refrigerant economizer
制冷剂节热器是一种有效的技术，它利用制冷剂的相变来节能或在外部温度条件许可的时候提供自然冷却。这些系统使用常见的基于压缩机的技术和泵在系统中循环制冷剂以提供冷却。它们可以在不消耗水的情况下提供非常高效的冷却，并且由于其简单的设计而具有最佳的10年总拥有成本。
Refrigerant economizer is an efficient technology that leverages the phase change of refrigerants to economize or provide freecooling during times when the temperatures outside will support it. These systems use common compressor-based technology and pumps to circulate the refrigerant through the system to provide cooling. They can provide very efficient cooling without the use of water and have the best 10-year total cost of ownership because of their simple design.
芯片级液冷
Liquid cooling – direct-to-chip
芯片级液冷使用水或制冷剂直接接触芯片从而带走服务器中发热量最大的器件产生的热量。芯片级的冷却吸收大约80%的热量到液体中，而剩余的热量通过数据中心的空气冷却系统进行冷却。这一技术不仅是冷却设备的一种有效手段，而且降低了设备内部的风扇耗能，从而提高了运行效率。
Liquid cooling – direct-to-chip takes water or refrigerant directly to the chip to take the heat off the hottest components within the server. Direct-to-chip cooling extracts about 80% of the heat into the liquid while the remaining heat is cooled through air cooling systems in the data center. This technique is not only an efficient means of cooling equipment but it also reduces the fan energy within the equipment thus improving operational efficiency.
浸没式液冷
Liquid cooling – immersion cooling
浸没式液冷是一种相对较新的冷却技术，它是将计算机设备浸没在非导电液体中，通过与发热器件直接接触的液体循环，并经过冷却的热交换器带走热量。这种方法相对于芯片级液冷的好处是所有服务器产生的热量都被液体捕获，因此可以省掉服务器中的风扇。
Liquid cooling – immersion cooling is a relatively new cooling technique that submerges computer equipment in a non-conductive fluid. Heat is removed by circulating liquid into direct contact with hot components then through cool heat exchangers. The benefit of this approach over direct-to-chip liquid cooling is that all of the server heat is captured by the liquid so fans can be eliminated from the servers.
拥抱变化
Bracing for Change
平衡电力效率和用水量对于每个数据中心都是一个独一无二的过程，会受到持续的技术变革的影响。综合考虑当地运营成本、可持续性目标、法规和社区影响，运营商应该寻找能够提供PUE和WUE最佳组合的解决方案。场站规模是做出正确选择的重要因素，因为没有一种适合所有因素的场站规模。向超大型运营商寻求想法，因为他们有动力在这个领域做一些最具创新性的工作。
Balancing power efficiency and water usage is a process that is unique to each data center and is affected by continual technology change. Operators should look for solutions that provide the best combination of PUE and WUE given local operating costs sustainability objectives regulations and community considerations. Facility size is an important factor in making the right choice as there is no one size that fits all. Look to hyperscale operators for ideas as they are motivated to do some of the most innovative work in this area.
也要关注硬件设计和AI领域的新趋势。器件制造商一直在寻找降低功耗的方法，例如降低处理器循环周期或在低活跃期关闭某些器件。作为人工操作的补充，基于人工智能的控制将通过更大规模地检测基础设施并调整工作负载以优化能源的使用。
Also keep an eye on emerging trends in hardware design and AI. Component makers are constantly looking for ways to reduce power consumption such as scaling back processor cycles or shutting down components during periods of low activity. AI-based controls will complement human operators by monitoring infrastructure at a much larger scale and adjusting workloads to optimize energy use.
边缘计算的增长趋势也许能够进一步优化效率，将一些运算转移至边缘侧的风冷服务器同时减轻网络和数据中心的负载。
The growing trend toward edge computing may enable further efficiencies by moving some processing to air-cooled servers in the field and relieving the load on networks and data centers.
运营商还应关注智能电网的发展以及由此带来的套现他们的UPS基础设施的机会，同时简化实现可持续性供电的路径。电力和用水效率领域是不断创新的领域，这将成为数据中心运营中最令人兴奋的方向之一。
Operators should also monitor the development of the Smart Grid and the opportunities it will present to monetize their UPS infrastructure while simplifying the path toward sustainable power generation. The field of power and water efficiency is the scene of constant innovation making it one of the most exciting dimensions of data center operations.
深 知 社
翻译：
沈文伟
威图电子机械技术（上海）有限公司 数据中心事业部 产品经理
DKV（DeepKnowledge Volunteer）计划成员
校对：
阮晨鹏
施耐德电气关键电源事业部开放式架构 资深设计师
DKV（DeepKnowledge Volunteer）计划成员
公众号声明：
本文并非原文官方认可的中文版本，仅供国内读者学习参考，不得用于任何商业用途，文章内容请以英文原版为准，本文不代表深知社观点。中文版未经公众号DeepKnowledge书面授权，请勿转载。
