|
什么是数据中心
What is a data center?
数据中心是对世界上的数据进行处理、组织及存储的场所。
Data centers are where the world’s data is processed, organized and stored.
很多数据中心设立在类似于仓库的建筑里，它们是一些功能性建筑，尽管有些建筑比其他建筑更具有美感，但它们几乎不需要进行物理访问。
Many data centers are located in buildings similar to warehouses. They are functional buildings that require little physical access - though some are more aesthetically pleasing than others.
由于它们往往是大型的，所以它们经常被建在广阔、土地便宜的场所，如美国的内华达州，但正如我们将在下面，以及其他文章中进一步探讨的那样，靠近其他运营商和网络供应商，以及获得廉价能源和政府补贴等因素也一定程度上影响着数据中心的选址。
Because they tend to be large, they are often built in vast open spaces, where land is cheap, like the US state of Nevada, but as we will explore further below - and in other articles - factors like proximity to other operators and network providers as well as access to cheap energy and government subsidies can play a part.
严格来讲，数据中心可以小到一个办公单元的柜子，但通常它们都比较大，包含成千上万的服务器或计算机通过物理或数字网络相互连接，也会连接到互联网。
Strictly speaking, a data center can be as small as a cupboard in an office unit, but typically they are larger, contain thousands of servers, or computers, connected to one another - and often, but not always, to the Internet - via physical or digital networks.
Citadel，隶属于一家名为Switch的公司，位于美国内华达州世界最大的商业园区Tahoe Reno工业中心，是世界上最大的数据中心园区之一。
One of the world’s largest data center campuses, The Citadel, located in the Tahoe Reno Industrial Center, the world's largest business park, in the US state of Nevada, is owned by a company called Switch.
完全建成后，Citadel将跨越几乎800万平方英尺（超过74万平方米），尽管一个或几个数据中心的规模和价值更多的时候是以其电力消耗方面的能力来衡量，因为这决定了它能容纳多少IT设备。Citadel在完全开发后，容量至少是650MW，这样的话，该站点将成为世界上最大的数据中心。但是，目前这个头衔仍属于Facebook/Meta。
At full build out, Citadel will span almost eight million square feet (that’s just over 740,000 square meters) - though the scale and value of one or several data centers is more often measured by its capacity in terms of electrical consumption, as that determines how much IT equipment it can accommodate. In The Citadel’s case, that’s 650MW - at least once it is fully developed. Should that happen, the site will be the world's largest data center - but currently, that title belongs to Facebook/Meta.
The backbone of the Internet
政府和银行使用数据中心来安全地进行存储和交换数据；研究机构用来压缩、分析和解析大量的信息；游戏公司依靠数据中心渲染图形并允许全球范围内的玩家进行在线游戏。
Governments and banks use data centers to store and exchange data securely; research organizations use them to crunch, analyze and interpret huge amounts of information; gaming companies rely on data centers to render graphics and allow users to play online around the globe.
军队依靠数据中心为武器系统提供动力；有了基础设施来处理绘制和解决复杂问题所需的大量数据，目前的人工智能是有进步空间的。
The military relies on data centers to power weapons systems; current advances in artificial intelligence are possible because the infrastructure is there to process the masses of data required to map out and resolve complex problems.
数据中心是功能性建筑，尽管有些建筑比其他建筑更具有美感，但它们几乎不需要进行物理访问。
Data centers are functional buildings which require little physical access - though some are more aesthetically pleasing than others.
每当你进行网络查询时（即在浏览器中输入网站名称、使用搜索引擎、点击链接、扫描二维码），你都在直接或间接和数据中心进行互动。
Whenever you make a web query (i.e., type in the name of a website into a browser, use a search engine, press a link, scan a QR code), you are interacting, either directly or indirectly, with a data center.
加密货币部门依靠同样的数字基础设施进行采矿，虽然它们通常被称为服务器农场，或加密货币农场，但它们本质上是数据中心，配备了高性能计算机。
The cryptocurrency sector relies on the same digital infrastructure to carry out mining. Though they are often called server farms, or crypto farms, they are essentially data centers, equipped with high-performance computers.
只要提供给人们接入点，存储在数据中心的信息就可以从任何地方访问。
Providing one has access to the networks they are connected to, information stored in data centers can be accessed from anywhere.
你离任何给定的数据中心有多近，以及它连接到什么网络，都会影响信息在网络的不同节点之间传播得时间-这被称为延迟。
How close you are to any given data center, as well as what networks it is connected to, affects the time it takes for information to travel between different nodes of the network - this is known as latency.
可以说，服务器室是大楼里最重要的房间，因为所有计算机都放在那里——在大型的垂直机架上，通过以太网连接到互联网。
Arguably the most important room in the building, the server room is where all of the computers are kept - in large, vertical racks, connected to the Internet via Ethernet connections.
一个标准的机架外壳是19英寸宽，通常包含多达42个服务器单元，用“U”来表示服务器单元的大小，通常为1U或2U。除了服务器，机架还包含存储系统、网络设备（如交换机和路由器）、电缆管理附件、配电装置（PDU）和气流管理系统。
A standard rack enclosure is 19 inches wide and normally contains up to 42 server units - shortened to U to describe the size of a server unit, usually 1U or 2U. As well as servers, racks contain storage systems, networking equipment such as switches and routers, cable management accessories, power distribution units (PDUs) and airflow management systems.
和家用电脑一样，服务器会产生大量的热量，这在摆满服务器的房间里会成为问题，这就是为什么数据中心运营商试图使用计算机房空调（CRAC）装置来保持服务器房间尽可能的凉爽；将机架布置成冷热通道配置，冷空气入口朝向一侧，热空气排放朝向另一侧；以及依靠空气循环或液体浸没技术的机架冷却单元。
Just like home computers, servers generate a lot of heat, which can become problematic in rooms full of them, which is why data center operators try to keep server rooms as cool as possible using computer room air conditioning (CRAC) units; laying out racks in hot and cold aisle configurations, with cold air intakes facing one way and hot air exhausts facing the other; and in-rack-cooling units, which rely on air circulation or liquid immersion technologies.
保持特定的湿度水平和机架内洁净度也很重要，因为灰尘和污垢的积累会影响设备的可靠性，还有可能造成短路。
Maintaining particular humidity levels and cleanliness within the racks is important too, as an accumulation of dust and dirt can affect equipment reliability, potentially causing short circuits.
大多数的数据中心由电网供电，能源被输送到数据中心内的配电装置（PDU），以保持对设备的稳定供应，因为电力出现任何差错都会导致停机，也就是所谓的宕机，反过来又有硬件损坏、数据丢失和设备执行的任何操作中断的风险。
Most data centers are powered by the grid, with energy fed to power distribution units (PDUs) within the data center to keep the supply to the equipment stable, as any loss of power can result in downtime - otherwise known as an outage - which in turn risks hardware damage, data loss and disruption to whatever operation the equipment is performing.
出于同样的原因，数据中心配备了备用电源系统，如发电机和不间断电源（UPS）。
For the same reason, data centers are equipped with backup power systems such as generators and uninterrupted power supplies (UPS).
随着电力、财务和环境成本的持续上升，公司正越来越多地寻求更多节能和碳密度更低的方式为其数据中心供电，如太阳能、风能，或采取其他有效措施，如使用外部空气来冷却IT设备，或投入耗电更少的硬件。
As the cost of power - financial and environmental - continues to rise, companies are increasingly seeking more energy-efficient and less carbon intensive ways of powering their data centers, such as solar or wind warms, or making other efficiencies, like using outside air to cool IT equipment, or investing in less power-draining hardware.
企业通常使用多个数据中心，有一个被称为“灾难恢复”站点的备份位置，在独立的服务器上存储其数据的副本，以避免在主站点断电或火灾的情况下丢失信息。
Businesses typically use multiple data centers, with a back-up location, known as a ‘disaster recovery’ site, which store a copy of their data on separate servers to avoid losing information in the case of, say, a power cut, or a fire at the primary site.
这些数据中心都在不同的地点，有足够远的距离以免同时遭受相同的灾难，但又足够近，可以与另一地点进行物理连接。
These are at a different location, far enough away to avoid being struck by the same disaster at the same time, but close enough that they can be physically connected to the other site.
对数据中心的访问往往受到严格的安全措施的限制，因为它们包含昂贵的设备，有时还有敏感或专用数据。
Access to data centers is often limited by tight security measures, as they contain expensive equipment and sometimes sensitive or proprietary data.
物理安全可能是指监控摄像头、警卫和密码，或者甚至只有生物识别才能进入。虚拟安全，是基于软件的，是防止网络攻击或其他试图入侵或危害设备设施内的数据的行为。
Physical security might mean surveillance cameras, guards, and passcode or even biometric access only. Virtual security, to protect against cyberattacks or other attempts to hack into or jeopardize the data contained within a facility, is software-based.
分布式拒绝服务（DDoS）是一种常见的数据中心系统攻击，攻击者用流量淹没这些系统，以使其失去功能，并使目标网站或服务离线。
Distributed-denial-of-service (DDoS) is a common attack on data center systems, whereby an attacker floods them with traffic in order to disable them and bring a targeted website or service offline.
攻击的目标是谁的系统，决定了它的影响有多严重：2016年，对安全公司Dyn服务器的一次重大DDoS攻击，对广泛使用的亚马逊网络服务（AWS）、Spotify和Twitter造成了重大破坏。
Whose systems are targeted in an attack decides how serious an impact it has: in 2016, a major DDoS attack on security firm Dyn’s servers caused major disruption to the widely-used Amazon Web Services (AWS), Spotify, and Twitter.
这种攻击也可能是国家支持的，这表明为了国家安全，有必要建立强大的规避系统。在2022年3月俄罗斯入侵乌克兰之前，该国最大的商业银行和国防部的网站都受到了DDoS的攻击-尽管关方都没有将其归咎于任何特定的团体或国家。
Such attacks can be state-sponsored, too, revealing the necessity for robust mitigation systems for the sake of national security. Before Russia invaded Ukraine in March 2022, the country’s largest commercial bank and defense ministry websites were both hit by a DDoS attack - though neither were officially attributed to any particular group or nation.
为防止严重的数据损失，数据中心在机房内外都配备了灭火系统。传统的泡沫或水基灭火器会导电，因此运营商更青睐于气体灭火系统，它对IT系统的损害最小。
To prevent serious data loss, data centers are equipped with fire suppression systems within and outside of server rooms. Conventional foam or water-based fire extinguishers conduct electricity, so operators favor gas suppression systems, which cause minimal damage to IT systems.
Different types of data centers
This is when companies decide to store, manage and protect their own IT equipment.
反过来说，这并不一定意味着数据中心位于公司总部或他们的办公室，但是他们管理这些设备。
Counterintuitively, this doesn’t necessarily mean the data center is located, say, at a company’s headquarters or in their offices, but that they own the equipment.
一些公司选择将其IT系统放在与其他企业共享的设施中，将网站的日常运行，比如维保和安全外包给第三方。
Some companies choose to place their IT systems in facilities shared with other businesses, outsourcing the day-to-day running of the site, like maintenance and security to a third party.
硬件是他们自己的，但是电力、冷却和与网络服务供应商的连接，通常由主机托管供应商保证。
The hardware is their own, but power, cooling, and connectivity to network service providers is usually ensured by the colocation provider.
供应商只将设施的大块或全部租给单一客户，是主机托管的一种形式。对于主机设施托管何时成为“批发”形式，没有行业标准。
A form of colocation where the provider only leases large blocks, or the entirety, of a facility to a single customer. There is no industry standard for when a colocation facility becomes 'wholesale.'
云数据中心与主机托管站点不同，客户数据存储在属于云供应商的系统上，信息被虚拟管理，而不是将每台服务器分配给特定用户。
Cloud data centers are different to colocation sites in that customer data is stored on systems belonging to the cloud provider, and the information is managed virtually rather than each server being allocated to a particular client.
对于希望以最小的投入来存储、处理和备份数据的企业来说，通常只需为他们所使用的数据付费，云计算通常是首选方案。
For businesses wishing to store, process and back up their data with minimal input, often only paying for what they use, cloud computing is often the preferred option.
超大规模数据中心是由有大量数据处理需求的公司所拥有或租赁的，这是因为他们自己监督着大部分互联网，或者因为他们为第三方管理着技术系统。仅举几例，比如Facebook/Meta，谷歌, 苹果, 亚马逊, IBM和微软。
Hyperscale data centers are owned or leased by companies with massive data processing requirements, either because they oversee massive chunks of the Internet themselves, or because they manage technology systems for third parties. Think Facebook/Meta, Google, Apple, Amazon, IBM, and Microsoft, to name a few.
在超大规模设施中，整个建筑通常被交给一家公司控制-比如谷歌搜索、Meta社交、亚马逊AWS云。这就形成了主要的规模经济，并允许超大规模运营商推动技术边界和试验新方法。
In hyperscale facilities, the entire building is typically given over to monolithic applications under one company's control - say Google's search, Meta's social media, Amazon's AWS cloud. This makes for major economies of scale, and allows hyperscale operators to push the technology boundaries and experiment with new approaches.
由于连接已经成为现代生活的必需品，从智能手机的使用到为无人驾驶或城市面部识别系统提供所需的即时传输数据，一些公司正在寻找将处理能力带到更接近需要的地方，即所谓的“网络边缘”，以减少数据传输到设施和返回目的地所需的时间，这也被称为“延迟”。
As connectivity has become a staple of modern living, from smartphone use to the instant data transmission required to power driverless cars or operate city-wide facial recognition systems, some companies are seeking to bring processing power closer to where it is needed, to ‘the network edge,' as it is known, to reduce the time it takes for data to travel to a facility and back to where it is needed - which, again, is known as latency.
数据中心的前身是最早的大型计算机，始于英国和美国陆军弹道研究实验室的电子数字积分器和计算机（ENIAC），这个实验室建于20世纪40年代，用于计算火炮密码。
The precursors of data centers were the first large computers, starting with developments in the UK, and the U.S. Army Ballistic Research Laboratory’s Electronic Numerical Integrator And Computer (ENIAC), built in the 1940s to calculate artillery codes.
冷战期间，信息处理、存储和加密被视为国防系统发展的重要内容，并对商业主机的最终诞生起到了重要作用。
Information processing, storage and encryption were seen as important in the development of defense systems during the Cold War, and were instrumental in the eventual creation of the commercial mainframe.
20世纪50年代，晶体管取代了电子学中的真空管，最终创造出了第一个集成电路，计算机的规模和数量随后迅速增加。起初，主机在本地保存和处理数据，没有像今天这样通过网络共享数据的能力。另外，这些设备又笨重又昂贵，所以对于那些要求以数字方式存储数据的公司来说，许多公司更倾向于保留纸质记录。由于在自己的系统中存储数据是不划算的，于是他们采用共享主机的方式。
In the 1950s, transistors replaced vacuum tubes in electronics, culminating in the creation of the first integrated circuit, and the size and number of computers increased rapidly. At first, mainframes held and processed data locally, without the ability to share it across a network like they do today. Plus, the equipment was bulky and expensive, so for those companies that required that their data be stored digitally - with many preferring to keep paper records - storing it on their own systems was not viable, and mainframes were shared.
但是，由于摩尔定律准确地预测了集成电路上的元件成本将随晶体管的增加而成比例地降低，20世纪80年代和90年代，微处理器的出现使处理能力（相对）便宜，所以更多的公司开始购置属于自己的硬件。
But as Moore’s Law accurately predicted that the cost of components on an integrated circuit would be reduced proportionately to how many transistors were added to it, the advent of microprocessors of the 1980s and 1990s made processing power (relatively) cheap, so saw more companies purchasing their own hardware.
网络使计算机能够跨越遥远的距离进行连接——然而矛盾的是，这导致了临近集群的发展。
Networks enabled computers to be connected across great distances - and paradoxically, this lead to the development of clusters located close to each other.
像主机托管、云计算和混合（企业内部和云计算的混合）这样的解决方案是后来才出现的，因为扩展系统的必要性越来越大，而专门从事云计算服务的公司在熟练程度和能力方面都有所提高。
Solutions like colocation, cloud, and hybrid (a mix of on-premise and cloud computing) came later, as the necessity to scale systems grew, and companies specializing in cloud provision increased in both proficiency and capacity.
网络使计算机能够跨越遥远的距离进行连接——然而矛盾的是，这导致了临近集群的发展。
Networks enabled computers to be connected across great distances - and paradoxically, this lead to the development of clusters located close to each other.
计算机开始集群发展，数据中心也随之出现。20世纪60年代，第一个互联的数据包共享网络“Arpanet”在高级研究计划局（ARPA）创建，这是一个位于弗吉尼亚州阿灵顿的政府机构。从Arpanet开始，互联网不断发展，第一个互联网交换中心MAE-East于1998年创建，也位于弗吉尼亚州北部。
Computers began to cluster, and data centers emerged. In the 1960s. the first interconnected packet sharing network "Arpanet" was created at the Advanced Research Projects Agency (ARPA), a US government agency based in Arlington, Virginia. From Arpanet, the Internet evolved, and the first Internet exchange MAE-East, was created in 1998, also in Northern Virginia.
随着服务供应商的出现，他们从靠近交换机的位置中获益。AOL和其它位于Loudon县的Ashburn，随着其他公司的涌入，变成了为人所知的“数据中心港”。第一个大型主机托管服务商Equinix在那里拥有大型设施，其他供应商在低税率和廉价土地的鼓励下纷纷进驻。
As service providers emerged, they benefited from being located close to exchanges. AOL and others located in Ashburn, Loudoun County, which became known as "Data Center Alley" as others piled in. Equinix, the first large colocation provider had large facilities there, and others moved in, encouraged by low taxes and cheap land.
Legal and geopolitical factors
除了决定企业在数据中心领域运营的商业驱动力之外，数据的存储和处理地点具有法律意义-有时是政治意义，其对该行业起着重要作用。例如，在欧盟，如果在境外存储欧盟公民的数据可能违反欧盟的《通用数据保护条例》，也就是非法的。
Beyond the commercial driving forces that decide where businesses operate in the data center space, where data is stored and processed has legal - and sometimes political implications which play a major role for the industry. In the EU, for example, it is illegal to store data belonging to the Union’s citizens outside of its borders if this might breach the EU's General Data Protection Regulation.
在其他地方，比如美国法律规定，如果政府认为，因调查需要，政府可以要求公司交出他们的数据（或者确切讲，是他们的客户数据）。
Elsewhere, like in the US, the law states that the government can demand that companies hand over their data, (or indeed, their customers’ data), should it be deemed relevant to an investigation.
在任何特定国家，谁拥有和运营数据中心也是一个争论的焦点：例如，中华人民共和国只允许外国公司与中国公司以合作的方式建设和运营数据中心。
Who owns and operates data centers in any given country can be a source of contention, too: for instance, the People’s Republic of China only allows foreign companies to build and operate data centers there if they partner with a Chinese company.
由于数据落地相关法律的改变，规定iCloud服务必须在国内提供，苹果公司出于对安全的考虑，长期以来一直拒绝在中国安营扎寨，最近投资10亿美元，与当地服务供应商贵州云合作，在中国建立了一个数据中心园区。
Because of a change in data residency laws stipulating that iCloud services must be delivered domestically, Apple, who out of concerns for security had long resisted setting up camp in China, recently invested $1bn in building a data center campus there in partnership with local service provider, Guizhou Cloud.
How much is the data center industry worth?
数据中心行业是世界经济的重要组成部分，因为它直接和间接地创造了就业机会，而且是技术服务行业的基础。20多年来，数据中心房地产市场一直在稳步增长，并受到Covid-19大流行的支持而非阻碍，因为大多数公司都加强了他们的数字商务能力并严重依赖在线会议系统。
The data center industry is an important part of the world economy, because it generates direct and indirect employment, and is the basis for the technology services industry. The data center real estate market has grown steadily for more than two decades, and was bolstered rather than hindered by the Covid-19 pandemic, when most companies ramped up their digital commerce capabilities and relied heavily on online conferencing systems.
根据技术研究和咨询公司Gartner的数据，2022年全球数据中心系统的支出将达到2260亿美元，而对数据中心容量的需求将达到历史新高。
According to technological research and consultancy firm, Gartner, global spending on data center systems will hit $226 billion in 2022, while the demand for data center capacity will reach an all-time high.
数据中心被视为有益于地方（有时是国家）经济，不是因为他们交了很多税，事实上，美国的一些州试图以大量的减税措施来吸引运营商，也不是因为他们直接雇佣了很多人（尽管他们在施工期间可以代表相当有利可图的合同），而是因为他们为该地区吸引了投资。
Data centers are seen as a benefit to local (and sometimes national) economies, not because they pay a lot of tax - in fact, some US states seek to attract operators with generous tax cuts - or because they directly employ a lot of people (although they can represent fairly lucrative contracts during construction), but because they attract investment to the area.
数据中心的房地产投资市场是巨大的，疫情期间，随着曾经有利可图的办公空间价值的下降，与其他类型的房地产相比，其增长速度不成比例。
The data center real estate investment market is huge, and, during the pandemic, as the value of once-lucrative office space fell, grew out of proportion compared with other types of real estate.
数据中心的交易可以达到天文数字：去年，上市房地产投资信托基金（REIT）American Tower以101亿美元收购了CoreSite的25个美国数据中心组合。
The buying and selling of data centers can reach astronomical sums: last year, listed real estate investment trust (REIT) American Tower bought up CoreSite’s portfolio of 25 US data centers for $10.1 billion.
Where are the data centers?
美国的数据中心远远多于其他国家，集中在东海岸的纽约、弗吉尼亚北部和亚特兰大；西海岸的旧金山和洛杉矶，以及中西部的芝加哥等地区。据房地产公司CBRE估计，大约3350MW的使用量用在了主要市场（在较小的枢纽有更多的容量）。据CBRE统计，另外还有730MW正在建设中。
The US has more data centers than any other country by a long way, concentrated in areas such as New York, northern Virginia, and Atlanta on the East Coast; San Francisco and Los Angeles on the West Coast, and Chicago in the Midwest. Real estate firm CBRE reckons there's around 3,350MW in use in primary markets (with more capacity in lesser hubs). There's another 730MW under construction, according to CBRE.
在欧洲、中东和非洲（EMEA），数据中心往往集中在互联网交易所所在地，最大的市场是法兰克福、伦敦、阿姆斯特丹和巴黎（被称为FLAPs），预计2022年总供应量将达到2200MW。在这个区域内，非洲是最小的一部分，但正在迅速扩大。
In Europe the Middle East and Africa (EMEA), data centers tend to be concentrated where the Internet Exchanges are, the biggest markets being Frankfurt, London, Amsterdam, and Paris (known as the FLAPs), with total supply predicted to reach 2,200MW in 2022. Within this region, Africa is the smallest part, but is expanding rapidly.
在亚太（APAC）地区，中国、澳大利亚和日本的数据中心数量遥遥领先，而且市场正在蓬勃发展：根据商业地产公司CBRE的统计，2021年，该地区数据中心的直接投资总额为48亿美元，比前一年增长了100%。CBRE认为，亚太地区大约有2000MW的在线容量，一些市场将在未来几年内增加近一倍。
In the Asia and Pacific (APAC) region, China, Australia and Japan lead the way in terms of the number of data centers located there, and the market is thriving: according to commercial real estate company CBRE, in 2021, direct investment in data centers in the region totaled $4.8 billion, a 100 percent increase on the previous year. CBRE reckons there's around 2000MW online in APAC, with some markets set to nearly double in the next couple of years.
拉丁美洲（LATAM）的数据中心市场要小得多，但也在增长。分析师不太可能列出该地区得具体数字，但据估计，目前的规模约为200MW，每年增长7%。
The data center market in Latin America (LATAM) is considerably smaller, but growing. Analysts are less likely to break out specific figures for the region, but estimates put the current size at around 200MW, growing at seven percent per year.
DKV（DeepKnowledge Volunteer）计划成员
本文来源于DCD，并非官方认可的中文版本，仅供读者学习参考，不得用于任何商业用途，文章内容请以英文原版为准（英文原版请点击“阅读原文”），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。
|