IDCNOVA feature: February 4,2019,the last day of Chinese Lunar Year of Dog.The annual spring festival gala is live on China’s national television to the whole country and the world.Hundreds of millions of viewers participated in the online hongbao(red carpet)crush instantly.As the sponsor,Baidu delivered a perfect answer,even encountering with 50 million times per second at peak time.No downtime.Alibaba and Tencent both encountered downtime as the major sponsor in the previous years.
Baidu Yangquan Cloud Computing Center crew supporting the spring festival gala live
Why Baidu did it this time?The answer is cloud computing capability.As a tech giant well-known across the globe,Baidu launched dozens of databases.And one of them is always attracting and it actually acted as the key role to support the spring festival gala broadcasting early 2019.That is Baidu Yangquan Cloud Computing Center.
“the largest single unit data center in Asia”,“AI cutting edge”“future model”,located in Northern China’s Shanxi Province,Yangquan Cloud Computing Center enjoyed lots of good names for its high performance,enormous investment and excellent design and most important,the world advanced technology.
Related: Baidu Cloud Computing Center claims power conservation affordable to 1,000 databases
Thus,even many outlets pressed reports on it home and abroad,this site is still appealing to so many people.And here we approach there closer again and clearly for a all-round observation,on its inner structure,theory,detailed configuration,operational ideology,executive insights and so on.AI everywhere is the stunning impression conveyed through tour.
Enormous Data storage
Baidu Yangquan Cloud Computing Center is scheduled to install over 280,000 servers,at present,more than 150,000 servers had been installed,with over 3 million CPU,which supported the spring festival gala live successfully and now are serving over 20 products,involving 100 million customers and more than 1 million enterprise clients.
In terms of data storage,it’s 6EB in this center,which is as 300,000 times as the total volume of collections in the China National Library.
Designed with T3+standard in 2011,Baidu Yangquan Cloud Computing Center was launched for construction since 2012.Over self-developed 400 patents of data center,network and server had been adopted in the available stage I,occupying 120,000 square meters of floor space.
Zhang Binghua,systematic director,Baidu
Lower and lower PUE
The annual PUE of Baidu Yangquan Cloud Computing Center was 1.1 in 2017 and it decreased to 1.09 in 2018.In the monitoring hall,the real-time indicators on display covers the operational status including PUE,load,IT service and capability,machine room electricity and heat.On the tour day,the single module unit with lowest PUE shows 1.062 and the annual PUE for 8 module units was 1.112.
Monitoring hall,Baidu Yangquan Cloud Computing Center
The annual PUE of China in 2018 is 1.73,the PUE of Baidu Yangquan Cloud Computing Center is 88%higher,ranking No.1 across the country and equivalent with overseas leading players like Facebook and Google.Simply put,250 million kwh could be saved per year in Baidu Yangquan Cloud Computing Center,enabling to guarantee power consumption of 130,000 households.
According to Zhang Binghua,Systematic Director of Baidu,starting from power reduction in the cooling period,Baidu targeted electric architect and air conditioning altering,as well as high temperature server and optimization to lower PUE.
High efficient power supply to maximum
Referring to electric architect,Baidu is making best to the maximum by 5 categories of architects like AC plus UPS,AC plus UPS ECO mode,AC plus HTDC online and AC plus HTDC offline,and AC plus HTDC offline in particular,which is the first one in the world.
In 2016,Baidu introduced its self-developed cabinet power supply technique and distributed lithium battery to replace the conventional UPS,which is electricity distribution room free,saving 42 milllion kwh each year and reducing construction investment by 70%as well as 25%of floor space,which can be used for more server to be installed.
Natural cooling
Generally speaking,data center adopts elevated floor for cooling,while Baidu Yangquan Cloud Computing Center adopted non-elevated floor at first in China.Its self-developed OCU air conditioning terminal integrating thresholding algorithm,applying the air convection current and brand-new air distribution device,which solved the power dissipation at the air conditioning terminal.Comparing the four types of air conditioning terminals including comprehensive water cooling air conditioning,AHU,OCU and indirect wind side,the energy consumption of OCU is zero.
To simplify cooling system by innovative technology,even the climate in Yangquan is not obviously favorable for data center operation,Baidu made extremely optimization on free cooling that the total free cooling time period reaches 96%all the year.
Complete cabinet delivery
Another reason for non-elevated floor for Baidu Yangquan Cloud Computing Center is that it adopted complete cabinet delivery mode to design as earliest one in China,thus the elevator in the center is designed to bear 4 tons specifically.In addition,barrier-free delivery design was implemented through all rounds covering loading platform,corridor,elevator and machine room.
By this way,the deployment node has been determined and completed before that of the self-developed completed cabinet by Baidu,and the online testing can be easily started after the complete cabinet“insertion”installment after arrival at the the data center,which saves manpower and provides convenience in years.
Powerful computing platform
Becoming the strongest computing platform of Baidu,Baidu Yangquan Cloud Computing Center holds several self-developed computing systems including Scorpio Complete Cabinet Server,X-MAN super AI computing platform,“Cold Mountain”chilled storage server and so on,provides powerful computing capability for products of Baidu and clients.
Scorpio Complete Cabinet Server,Baidu Yangquan Cloud Computing Center
In the module 7 and 8,AI technology was achieved at data center.Baidu AI high density computing cluster,the first AI computing cluster,Baidu adopted heterogeneous computing like GPU at first and introduced deep learning and heterogeneous computing cluster,becoming a dynamic engine for the brain of Baidu.
AI computing cluster
The AI computing capability of X-Man system equals tens of generic servers.As the first GPU solution by liquid cooling domestically,X-Man 2.0 launched in 2017 has been into scale application,accomplishing cooler-free operation completely and supporting faster implementation of AI.
High speed lossless network
Super powerful computing capability needs configuration of high speed network.The network architect of Baidu is prestigious for most stable,fastest transmission and largest data volume.
Baidu Yangquan Cloud Computing Center adopts self-developed high-performance interchanger,providing largescale network throughput,supporting 10G,25G generic computing network access and 100G high speed lossless network interconnection.
For the data center internal,Baidu connects tens of thousands of servers into a super computer via broadband,low latency and lossless network,becoming the most powerful cutting edge of Baidu in the AI era.
For the data center external,Baidu constructed 3 latency cycles:the latency less than 30ms from data center to customer coverage,the latency less than 10ms from POP point to customer coverage,and the latency less than 2ms from CDN to customer coverage,ensure the complete coverage and access nearby for customer across the country.
During the tour,the redundancy design was stressed most frequently by the Baidu officials.The only target is just and always one,no downtime ever,not matter what kind of means,power distribution or cooling machine,complete cabinet power supply and business operation.By April 8 of 2019,Baidu Yangquan Cloud Computing Center has been safely operated for 1,744 days.
How to save energy?
Baidu utilized the spare room on the roof in the Yangquan Cloud Computing Center utmost,installed clean energy,photovoltaic power generation,solar power generatiuon accessed to the power grid,becoming the first one for photovoltaic power generation in China.
Photovoltaic power generation on the roof at Baidu Yangquan Cloud Computing Center
Wind power contributed 26 million kwh in 2017 at Baidu Yangquan Cloud Computing Center,accounting 16%of the total power consumption.In 2018,wind power reached 55 million kwh.
Besides electricity,Baidu Yangquan Cloud Computing Center also treated highly on water resources recycling.The recycled water annually is 480,000 tons,equals to that of 4,000 households in water consumption per year.
Sewage recycling system
Intelligent operation
Occupying 120,000 square meters of floor space,while there are only tens of staff responsible for operation and maintenance.
The O&P here has accomplished complete intelligence mode.As for the patrol inspection,staff at Baidu Yangquan Cloud Computing Center adopted electronic procedure by Pad to inspect and upload data,which are designated with up limit,down limit and alert.
Referring intelligent heat dissipation,Baidu introduced AI into data center,established deep learning mode,the cooling water machine will switch to cooling,pre-cooling and energy-saving modes as per the outdoor moisture,temperature and loading capability.
In addition,the intelligent precaution system will inform exact fault location for the following maintenance or repair as per the equipment status and abrasion.
Compiled by China IDCNOVA, retransmission prohibited without authorization.
For more information, please contact justin@idcnova.com
