在ST Telemedia GDC用于數據中心冷卻的AI試點
Inside ST Telemedia GDC’s AI Pilot for Data Center Cooling
Wylie Wong，Aug 10 2021
今年春天，ST Telemedia GDC啓動了一項試點，研究基于人工智能的系統是否可以通過比設施工程師更有效地管理冷卻系統來減少其新加坡數據中心的能耗。
This spring, ST Telemedia Global Data Centres launched a pilot to see whether an AI-based system could reduce energy use of the cooling system in one of its Singapore data centers by managing it more efficiently than human facilities engineers could.
STT GDC的CTO Daniel Pointon最近告訴DCK，該試點已經産生了一些預期的初步結果。根據初始模型，冷卻系統的能耗比以前降低了約10%。
And the pilot has already produced some promising early results, Daniel Pointon, STT GDC’s CTO, recently told DCK. The cooling system stands to use about 10 percent less energy than before, according to initial modeling.
Pointon說：“這是相當重要，我們並沒有要求客戶關閉服務器或使用更少的電力，我們只是更有效地做同樣的工作。”
That’s “pretty substantial, given that we are not asking customers to shut down servers or use less power,” Pointon said. “We are just doing the same job more efficiently.”
如果考慮到STT GDC的規模和在整個行業中部署類似解決方案的節能潛力，它會變得更加重要。這家總部位于新加坡的主機托管商是投資公司ST Telemedia的子公司，在亞洲和英國運營著130個數據中心。
It becomes a lot more substantial if you consider STT GDC’s scale and the potential for energy savings if a similar solution is deployed across its entire portfolio. The Singapore-based colocation provider, a subsidiary of the investment company ST Telemedia, says it operates 130 data centers in Asia and the UK.
測試人工智能在數據中心冷卻管理中潛力的試點是STT GDC進一步承諾將減少其整體碳足迹並到2030年實現碳中和的一部分。其最大客戶已做出了減少碳排放的承諾，在此推動下，爲了達到可觀的程度，越來越多的專業數據中心運營方采取了比以前更大膽的舉措來解決他們的碳足迹問題。
The pilot to test AI’s potential in data center cooling management is part of STT GDC’s broader pledge to reduce its overall carbon footprint and become carbon-neutral by 2030. Pushed to a substantial degree by their largest customers who have made ambitions carbon-reduction pledges of their own, more and more specialist data center providers have been taking bolder steps than before to address their carbon footprint.
AI可減少數據中心制冷能耗
AI Can Reduce Data Center Cooling Energy Use
人工智能在數據中心管理中的應用還爲時過早。僅少數公司有過嘗試，最著名的例子是谷歌使用其DeepMind子公司開發的人工智能算法來管理其超大規模計算設施中的制冷系統。一位谷歌高管在2018年告訴我們，該方法使數據中心冷卻系統的能源使用量減少了30%。
It’s very early days for AI in data center management. Few have tried it, but the most famous example is Google’s use of AI algorithms developed by its DeepMind subsidiary to manage cooling systems in its hyperscale computing facilities. A Google executive told us in 2018 that the approach resulted in a 30 percent reduction in energy use by data center cooling systems.
冷卻只是STT GDC人工智能實驗的開始。Pointon希望公司在未來研究使用人工智能進行預測性維護、容量管理、安全分析和自動化業務流程。
Cooling is just the start of STT GDC’s experimentation with AI. Pointon expects the company to investigate using AI for predictive maintenance, capacity management, security analytics, and automating business processes in the future.
STT GDC聘請瑞士工業自動化巨頭ABB在新加坡進行爲期12個月的試點。Pointon表示，在試點的第一階段，ABB正在收集數據、構建 AI 模型並開發控制邏輯。
STT GDC hired the Swiss industrial automation giant to conduct the pilot in Singapore over a 12-month period. In the first phase of the pilot, ABB is gathering data, building the AI model, and developing the control logic, Pointon said.
這家托管服務提供商之前安裝了數千個傳感器來測量其制冷系統的各部位，包括整個數據中心內的溫度、濕度和氣流——在機架前、管道系統中和屋頂。還有傳感器測量制冷系統的每個組件的能耗。
The colocation provider previously installed thousands of sensors measuring different aspects of its cooling systems, including temperature, humidity, and airflow throughout its data centers – in front of racks, in the ductwork, and on the rooftop. There are also sensors measuring energy use by each component of the cooling system.
“所有這些傳感器都爲我們提供了一個良機，可以真正利用人工智能的力量並有意義地利用所有這些信息，並幫助我們做出關于我們應該如何運行各種設備以獲得最佳結果的操作決策，”他說。
“All those sensors give us a great opportunity to really harness the power of AI and to make meaningful use of all that information and to help us make operational decisions about the way we should operate various pieces of equipment for optimal outcomes,” he said.
尚未將制冷管理交給AI
Not Handing Cooling Management Over to AI Just Yet
分析使操作員能夠微調數據中心控制系統的設定點。
The analytics enable operators to tweak set points of the data center control systems.
“例如，它可以對泵的運行速度和水流量進行微調，或者微調冷卻塔或精密空調內的風扇速度，”Pointon解釋說。“這是對一個大系統進行大量微調的組合，並根據不斷變化的情況實時動態進行這些調整節省開支。”
“For example, it can make slight adjustments to the speed the pumps run and how much water we push around, or slight adjustments to fan speeds in our cooling towers or inside our CRAC units,” Pointon explained. “Those combinations of a lot of small adjustments over a big system and making those adjustments dynamically in real time based on changing situations [can] unlock savings.”
在試點的第二階段，ABB將在新加坡數據中心運行人工智能模型。STT GDC 不會讓系統自動調整設定值。Pointon說，數據中心經理將分析AI的建議，並在試用期間手動進行更改。
In the second phase of the pilot ABB will run the AI model at the Singapore data center. STT GDC won’t let the system adjust set points automatically. The data center’s managers will analyze the suggestions from AI and manually make changes during the pilot, Pointon said.
他說：“我們不會立即放手，並且完全相信該試點可以爲我們經營業務。”“我們將查看人工智能提出的控制建議，然後從我們人類視角看是否可行，或者這是否會導致我們在其他目標如可靠性方面犯錯誤？然後去落實一些更有意義的建議。”
“We are not going to let it loose right away and put all the faith that this pilot can run our business for us,” he said. “We will look at the control suggestions that AI makes and then we will put a human lens on top of that to see [whether it] make sense, or will that cause us to trip up on our other objectives like reliability? And then [we will] go execute some of those more meaningful suggestions.”
如果試點成功，下一步STT GDC將在位于中國、印度、英國、新加坡、印度尼西亞、泰國和韓國的數據中心部署人工智能技術。Pointon說，隨著時間的推移，人工智能可以更多地集成到數據中心運營中，並從簡單地提出建議轉爲自動運行系統。眼下，他不想太超前，想先看看試運行的效果。
If the pilot is successful, the next step would be deployment of the AI technology throughout STT GDC’s data centers, located in China, India, the UK, Singapore, Indonesia, Thailand, and South Korea. Over time AI could become more integrated into data center operations and go from simply making suggestions to automatically running the systems, Pointon said.For now, he doesn’t want to get ahead of himself and wants to see the results of the pilot first.
“除了關注能源方面，我們還必須利用該試點做出一些判斷，例如這對可靠性、成本和爲客戶提供的服務水平意味著什麽。我不想預先判斷結果。我們將完成這個試點。目前它看起來非常有意義。”
“As well as looking at the energy side of it, we’ve got to use this pilot to make some judgments [such as], what does this mean for reliability, costs, and our service levels for our customers. I don’t want to prejudge the outcome. We are going to finish this pilot. But it’s looking really positive.”
DeepKnowledge
翻譯：
王磊
阿裏巴巴集團有限公司數據中心運維工程師
校對：
Eric
DKV（Deep Knowledge Volunteer）創始成員
公衆號聲明：
本文並非原文官方認可的中文版本，僅供國內讀者學習參考，不得用于任何商業用途，文章內容請以英文原版爲准。中文版未經公衆號DeepKnowledge書面授權，請勿轉載。