Halvor Bjerke是DigiPlex的首席運(yùn)營(yíng)官

由DCD中國(guó)授權(quán)DKV編譯并發(fā)表于DeepKnowledge微信公眾號(hào)。

作為一個(gè)評(píng)價(jià)指標(biāo)，電能使用效率(PUE)具有不可否認(rèn)的優(yōu)點(diǎn)。自2007年引入以來(lái)，它提供了一種簡(jiǎn)單的方法來(lái)評(píng)價(jià)數(shù)據(jù)中心效率的進(jìn)展。該參數(shù)是一個(gè)簡(jiǎn)單的數(shù)字，由數(shù)據(jù)中心的總能耗除以其IT設(shè)備的能耗，顯示了制冷、供電和其他因素所占用的其他能源。

由于該指標(biāo)簡(jiǎn)單可行，因此在整個(gè)行業(yè)中得到了長(zhǎng)足的改進(jìn)。從2007年到2018年，整個(gè)行業(yè)的平均PUE從2.5下降到1.58，這意味著能源損失減少了60%以上。

但幾乎所有這些進(jìn)步都是在前半期取得的。平均PUE在2013年已經(jīng)是1.65，到2018年只下降了0.07。然而在2019年5月Uptime研究所發(fā)布最新數(shù)據(jù)時(shí)卻上升了，目前該指數(shù)為1.67。那么，進(jìn)展是否已經(jīng)停滯，甚至開(kāi)始逆轉(zhuǎn)?從局外人的角度來(lái)看，仍然有大量的能源被“浪費(fèi)”，并伴隨著巨大的“碳成本”。

舉手之勞

可惜的是，似乎已經(jīng)取得了一些輕松的進(jìn)展。例如冷熱空氣分離、免費(fèi)冷卻、變頻風(fēng)扇控制和電量分配等措施在早期提供了巨大的進(jìn)步。如今，一切都取決于細(xì)節(jié)，并帶來(lái)了新的風(fēng)險(xiǎn)。依賴(lài)新風(fēng)系統(tǒng)使數(shù)據(jù)中心容易受到火災(zāi)等當(dāng)?shù)貧夂虻挠绊?對(duì)UPS的妥協(xié)讓需要對(duì)增加當(dāng)?shù)仉娋W(wǎng)的可靠性。對(duì)于一個(gè)依賴(lài)于單個(gè)數(shù)據(jù)中心的客戶來(lái)說(shuō)，這種做法可能很難被接受，對(duì)他們來(lái)說(shuō)，即使是一秒鐘的停機(jī)也可能造成災(zāi)難性的損失。

因此，盡管效率仍然至關(guān)重要，除大型集群和超大規(guī)模數(shù)據(jù)中心外，繼續(xù)專(zhuān)注于降低PUE可能會(huì)降低自身的收益。目前朝向混合解決辦法的方向，也正在削弱PUE作為單一指標(biāo)的效力。隨著服務(wù)器遷移到云上，剩余服務(wù)器機(jī)房的PUE會(huì)增加，雖然總體能耗在下降，但被更少的服務(wù)器所割裂。這種向公共云的遷移似乎推動(dòng)了2019年全球平均PUE的增長(zhǎng)。

真正的問(wèn)題是，PUE并沒(méi)有真正反映出其他類(lèi)型的效率提高，而這些效率提高在一個(gè)數(shù)據(jù)中心的業(yè)務(wù)環(huán)境狀況中扮演著越來(lái)越重要的角色。以熱回收為例，使用數(shù)據(jù)中心的IT設(shè)備產(chǎn)生的熱量為家庭和企業(yè)提供熱源，或通過(guò)熱泵直接發(fā)電。它有可能大幅減少能源浪費(fèi)，但所需的額外設(shè)備可以提高該中心的PUE評(píng)級(jí)。

當(dāng)然，在現(xiàn)實(shí)世界中，客戶很少基于單一因素做出決定。2018年12月，Supermicro發(fā)布的研究表明，59%的決策者認(rèn)為PUE很重要。但是，令人吃驚的是，58%的人甚至不知道他們的數(shù)據(jù)中心的PUE。研究還發(fā)現(xiàn)，只有9%的人將能源效率，作為其策略的首要因素，安全性、連接性和使用性能都被視為同等的，甚至對(duì)決策的影響更大。然而，由于PUE影響到運(yùn)營(yíng)的許多其他方面，從成本到聲譽(yù)，這個(gè)數(shù)字可能低估了效率對(duì)決策的實(shí)際影響。

對(duì)于數(shù)據(jù)中心和客戶來(lái)說(shuō)，效率始終是核心考慮因素——以減少成本和保護(hù)環(huán)境。隨著審查的增加，關(guān)鍵是要保留一個(gè)多樣的評(píng)價(jià)方法，而不是過(guò)度依賴(lài)某個(gè)單一的度量，應(yīng)考慮更廣泛的量度及影響，包括:

• 使用可再生能源。調(diào)查電力供應(yīng)的來(lái)源，但理想的是使用一個(gè)綠色的可再生能源供應(yīng)鏈，保證每千瓦電能消耗都是直接來(lái)自于可再生能源。

• 更好的設(shè)計(jì)。現(xiàn)代、模塊化和靈活、專(zhuān)業(yè)的的設(shè)施效率更高，可以定制化，以滿足精確的 KPI，以實(shí)現(xiàn)性能和效率。

• 專(zhuān)業(yè)。數(shù)據(jù)中心運(yùn)營(yíng)的商業(yè)和環(huán)境影響如此之大，需要由具有特定技能和知識(shí)的專(zhuān)門(mén)團(tuán)隊(duì)來(lái)管理，以便做出正確的決策。

• 更好的維護(hù)。這些專(zhuān)家團(tuán)隊(duì)將了解專(zhuān)業(yè)設(shè)備，并保持其在最佳水平上工作，從而節(jié)省資金并提高可持續(xù)性。

• 創(chuàng)新。在數(shù)據(jù)中心的世界里，沒(méi)有什么是靜止的。創(chuàng)新是持續(xù)的，每次升級(jí)都能提高效率，降低能源成本。確保您充分利用最佳創(chuàng)新，并是可持續(xù)發(fā)展持續(xù)進(jìn)步的基礎(chǔ)。

• 合作關(guān)系。每個(gè)客戶，每個(gè)負(fù)載，都是不同的。對(duì)于一些冗余將是至關(guān)重要的，對(duì)于其他人來(lái)說(shuō)，直接云連接——服務(wù)和靈活地適應(yīng)不斷變化的需求，有機(jī)會(huì)去提高創(chuàng)新能力。

任何指標(biāo)都必須視為路標(biāo)，而不是解決方案本身。盲目遵守任何單一措施都可能扭曲決策，最終阻礙這一重要領(lǐng)域的進(jìn)展。隨著監(jiān)察力度的加大，數(shù)據(jù)中心運(yùn)營(yíng)商及其業(yè)務(wù)客戶必須定期向自己，提出更深入的問(wèn)題，以揭示解決方案的真正效率，而不是依賴(lài)于像PUE這樣的單一“魔性”指標(biāo)。

這種對(duì)效率的實(shí)際效果更廣泛、更全面的看法將會(huì)擁有更好的商業(yè)、環(huán)境和聲譽(yù)。

英文原文

Halvor Bjerke, DigiPlex

Halvor Bjerke is chief operating officer at DigiPlex

Further sustainability gains may be lost through a narrow focus on PUE 

As a metric, Power Usage Efficiency - or PUE - has undeniable merits. Since its introduction in 2007, it has offered a simple way to chart progress in data center efficiency. The rating is a single figure, derived by dividing the data center’s total energy consumption by that of its IT equipment alone - revealing the excess taken up by cooling, secure power supply and other factors.

Being simple and seemingly transparent, use of the metric has led to a general improvement across the industry. Between 2007 and 2018, average PUE across the industry fell from 2.5 to 1.58, meaning that energy losses were reduced by more than 60 percent. 

But almost all of this progress was made over the first half of the period. Average PUE was already at 1.65 in 2013, falling only 0.07 to 2018 - and then rising when the latest data was published by the Uptime Institute in May 2019. It currently stands at 1.67. So, has progress stalled - or even begun to reverse? From an outsider’s point of view there’s still an awful lot of energy ‘wasted’ with a huge ‘carbon cost’ associated with it.

Low hanging fruit

Worryingly, it seems the easy gains have already been made. Measures like hot/cold air separation, free air cooling and better control over fans and power distribution offered big jumps early on. Now everything lies in the detail - and presents new risks. Reliance on external air leaves the center vulnerable to local events like fires; compromising on UPSs places tremendous faith in the reliability of the local grid. That faith might be hard to share for a client reliant on a single data center, for whom even a second of outage could prove catastrophically expensive.

So, while efficiency remains vital, continuing to concentrate on driving down PUE for its own sake might offer diminishing returns to all but the largest colocations and hyperscale centers. Current moves towards hybrid solutions are also undermining the efficacy of PUE as a single indicator. As servers are moved to the cloud, the PUE of remaining server-halls increases because overall energy consumption, although falling, is divided by fewer servers. It is this migration to the public cloud that seems to have driven 2019’s global rise in average PUE.

The real issue is that PUE doesn’t truly reflect other kinds of efficiency gain, which form an increasingly important part of a center’s environmental profile. Take heat return, using heat produced by the IT equipment in the data center either to provide heating for homes and businesses, or via heat pumps to generate power directly, as an example. It has the potential to substantially decrease energy waste, but the additional equipment needed could increase the center’s PUE rating.

In the real world, of course, clients rarely make a decision based on a single factor. In December 2018, Supermicro published research suggesting that 59 percent of decision-makers considered power efficiency important. But, strikingly, 58 percent did not even know their data center’s PUE. The research also found that only nine percent made power efficiency the primary factor in their strategy with security, connectivity and performance all seen as equal if not greater influences on decision making. However, because efficiency impacts so many other aspects of operations, from cost to reputation, this figure likely underplays the real impact of efficiency on decision-making.

For both data centers and clients, efficiency must remain a central consideration - to reduce both costs and environmental footprint. As scrutiny mounts, the key is to retain a sophisticated approach, without over-relying on any single metric. A wider range of metrics and impacts should be considered, including:

• Using renewable energy. Audit where electricity comes from, offsetting helps, but the ideal is an auditable energy supply chain that guarantees every kWh consumed has been directly generated from a renewable source.

• Better design. Modern, modular and flexible, purpose-built facilities are much more efficient and can be set up to meet precise KPIs for performance and efficiency.

• Expertise. The commercial and environmental impact of data center operations is so large they need to be managed by dedicated teams with the specific skills and knowledge to make good decisions.

• Better maintenance. These expert teams will understand specialist equipment and keep it working at optimum levels saving money and increasing sustainability.

• Innovation. Nothing stands still in the world of data centers. Innovation is ongoing, and each iteration offers enhancements to efficiency and reductions in carbon cost. Ensuring that you are capitalizing on the best innovations is fundamental to continued progress on sustainability.

• Partnership. Every customer, every load, is different. For some redundancy will be critical, for others direct cloud connection - advising and flexibly adapting to changing needs can create opportunities to create new efficiencies.

Any metric must be seen as a signpost rather than the solution itself. Blind adherence to any single measure can distort decision-making and ultimately impede progress in this essential area. As scrutiny intensifies, rather than relying on a single ‘magic bullet’ metric like PUE, data center operators, and their customers in the business, must routinely challenge themselves with deeper questions that uncover the real efficiency of their solutions.

This wider, more comprehensive view of real progress on efficiency will lead drive better commercial, environmental and reputational outcomes.

第十屆DCD>北京國(guó)際峰會(huì)活動(dòng)將于12月5號(hào)在北京舉辦。在本次大會(huì)中，我們將圍繞“超大規(guī)模數(shù)據(jù)中心快速建設(shè)”，“可持續(xù)發(fā)展及能源效率改善”，“人工智能與數(shù)據(jù)中心運(yùn)營(yíng)”等三大主題邀請(qǐng)超過(guò)1000位國(guó)內(nèi)外專(zhuān)家集聚一堂，就行業(yè)最新熱點(diǎn)話題展開(kāi)討論！ 

您想與國(guó)際數(shù)據(jù)中心服務(wù)商和投資方面對(duì)面嗎？