Shulou(Shulou.com)06/01 Report--

Guide: Zhu Hua, director of Tencent data center technology development center, vice chairman of data center technical committee of China Engineering Construction Standardization Association, leader of data center working group of China Communications Standardization Association Open data Center Committee, won the 2018 data Center Young Science and Technology Talent Award issued by China Engineering Construction Standardization Association. Zhu Hua delivered a speech at the 2018 data Center Annual Summit on November 30th. The following is the content of the speech.

Talking about the exploration of the next wind direction of the data center, I would like to take this opportunity to review the history of Tencent data center's exploration of technology development in the past decade.

One. Ten-year history of technology exploration in Tencent data center

1. High voltage DC with many birds with one stone

From 2008 to 2010, in order to solve the problem of power conversion efficiency, we launched the technical architecture transformation project of 240V HVDC instead of UPS on the basis of Mr. Zhao's laboratory of Yancheng Telecom. This is a process of reconstructing the low-voltage distribution architecture. At first we were struggling, but now BAT has millions of servers running on this distribution architecture.

Interestingly, we originally intended to save energy, but as a result, the greater benefits are the convenience of operation and maintenance and the reliability of operation. Throughout the industry, all data centers using this distribution architecture have never affected their business because of serious HVDC problems. In contrast, UPS distribution architecture often leads to heavy business losses due to frequent failures.

2. Taste the sweetness of standardization and production for the first time-- micromodule

From 2011 to 2012, we improved the architecture of the micromodule and solved the problems of standardization and partial structural production in the IT module. The micro-module has changed the engineering practice in the original IT module, subverted the previous data center construction interface and mode, and even broke the original pattern of the industry. Up to now, the annual production of micromodules is about 5000 sets.

3. Normandy model of ten thousand laws returning to clan

In 2013, Tencent's large-scale data center construction in Qingpu and Shenshan needs to standardize the methodology for the planning and design of large-scale (10-150000 server scale) data centers. So we launched an exploration project codenamed "Normandy Model". This methodology is a set of logical algorithms. It solves one of the most important problems-to explain the logic of the construction requirements of each data center, to make the requirements clear, to let the design institute draw clearly, to make the construction of the general contract clear, and for operators to take over. All projects from construction to mechanical and electrical construction are a unified standard. In this way, Party A, the design institute, the general contractor, the acceptance party and other relevant parties have the same understanding and direction, and regard the data center as a standard industrial product rather than a unique art product.

4. Refrigeration equipment should be dehydrated, deengineered, manufactured and localized.

In the early days, Tencent tried airside freecooling technology in Tianjin, which is more mature in the United States, but it is not satisfied in China: harmful substances in the air go directly into the computer room, wasting IT equipment. How can we make full use of the external cooling capacity without exposing the IT equipment directly? In 2015, we introduced China's first set of air-to-air heat transfer refrigeration equipment from GEA in Germany in Guiyang laboratory, and made a technical transformation to adapt to China, thus achieving great success. This technology exchanges heat rather than matter-it uses engineered polymer or metal materials to exchange heat between internal and external air, but the gas is isolated. Today, this kind of equipment can be made in China, which is as efficient as imported products, but the cost is completely halved. The industry has been applied on a large scale, and the whole industrial chain has sprung up. The success of the exploration of this technology goes far beyond energy conservation itself. Because this kind of indirect refrigeration is dehydrated and deengineered, it gives up the large-scale water-cooled unit, that is, gets rid of the complicated pipeline engineering. Our refrigeration equipment has finally embarked on the road of production!

5. The product of the overall standardization and production of the data center: T-block

The realization of the production of refrigeration equipment gives us the opportunity to explore the overall production of the data center. In 2016, we launched T-block technology, designed refrigeration square warehouse, IT square warehouse, electric power square warehouse, medium voltage square warehouse, and followed the whole industry chain. At this point, the data center can be built and delivered in a fully productive manner.

6. Normandy model re-evolution-Tbase model

In 2018, in order to cooperate with the large-scale self-construction of land acquisition across the country, we evolved the "Normandy Model" methodology into a new construction methodology for large and super-large campuses, the "Tbase Model", to deal with the planning and design of data centers with 300000 servers and above.

After ten years of exploration, we have always adhered to the road of modularization, standardization and production. This is a gradual process, accompanied by the follow-up and cultivation of the industrial chain. The road of reconstructing the traditional structure and subverting the traditional way of construction is full of inhumane hardships.

2. Future exploration direction of Tencent data center technology development

1. Higher requirements for data storage, calculation and transportation in the era of artificial intelligence.

So far, at the hardware level, we seem to have reached a self-breaking bottleneck and ceiling, but it is not. As a basic service industry, data center will face a new period of development-the artificial intelligence era of the Internet of everything. In the past years, the extremely hot development of cloud computing, big data and the Internet of things actually married for this artificial intelligence era of the Internet of everything and laid the foundation for the emergence of this era.

The realization of artificial intelligence depends on several basic conditions: the massive storage capacity of ①, the more advanced algorithm ③, the communication channel with large capacity and large bandwidth. Big data as the original basis of artificial intelligence, cloud computing for high-energy computing and storage, the Internet of things to provide ubiquitous data collection scenarios, 5G with very low latency for high-speed transmission.

2. The ducks with warm water in the river are the first to feel the smell of spring-Internet giants are facing the new era

The recent organizational restructuring of Ali and Tencent can be seen from the name: Ali's newly established BG name is followed by "intelligence". After the adjustment of Tencent Cloud, the name of BG has been added to "smart industry". Since then, the cloud no longer exists independently, but with application scenarios. The application scenario is to do increments, while in the past we mostly did inventory. Stock means the closure and transformation of the original IT, accompanied by a large number of transformation and migration hardships. After all, the amount of process is the amount of process, and the key to success or failure lies in the landing industry.

3. The stage of data center management software development: from facade and auxiliary management to production management.

Faced with new opportunities, the data center, as a branch of the industry, needs both hardware and software to realize the artificial intelligence data center of the Internet of everything. The hardware is ready for seven, seven, eight, eight. What about the software? Data center, what kind of intelligent brain do we really need? What are the prerequisites for unattended and artificial intelligence platforms and scenarios that can land? This is the next direction that we need to explore and practice.

Here, let's first review the past development. I have listed the three stages of the development of the data center management software platform, and to sum up, it has been very difficult in the past.

The first stage is in 2000-2010, when the typical representative products are moving rings and computer room 3D presentation, the total value of the market is 100-10 billion. Our Party An also did not pay attention to this piece, from the commonly used bidding mode, we can see that this is usually put in the total package. Manufacturers pay more attention to the cool 3D visual effects to win Party A's eyeball, but it is not practical. During this period, I can be summarized as the "facade system stage".

The second stage, from 2011 to now, the typical representative products of this period are relatively rich. DCOM, DCIM, DCSM and DCRM begin to control information from the perspective of equipment, people and process, and present it as data and reports for equipment health management, asset management, simple cost and capacity management. The total value of the market is 10-5 billion. Party An also began to prepare technical specifications, subcontract and bid independently. During this period, I can be summarized as "Auxiliary Management system stage".

The third stage is what we all need to explore and strive to achieve, combined with data center hardware products, its market value may exceed 100 billion. During this period, I summarized it as the "production management system stage". It will once again subvert most of the modes, habits, and organizational boundaries of operation and maintenance. Few people on duty, unattended, artificial intelligence will promote the data center operation and maintenance management system to the production management system, its importance can not be compared.

4. An industrial chain ready to go

Is the industrial chain fully prepared for the new era? First of all, let's take a look at the preparation of all kinds of sensors: in this era, we need a large number of sensors that are easy to deploy, for our eyes and ears, we need wireless and passive, you see, from temperature measurement, speed measurement, noise measurement, location measurement, pollution measurement and thermal imaging, and even a variety of controllers and actuators, the industry has opened its arms to welcome our exploration and innovation.

5. Unattended application scenarios: security, patrol, supervision and control of equipment

① Security:

Here we briefly list a few unattended application scenarios of artificial intelligence: we start from the manpower-intensive security personnel, ask for manpower from science and technology, and reduce the work of security personnel, such as seeing, touching, running errands, and so on. In advance, we use security robots, deploy a large number of high-definition video cameras, through active monitoring and defense, delineate the monitoring area in advance, define the monitoring list of important items, and control the key fire prevention areas. In the matter, we predict the human behavior through the algorithm, and warn the suspicious people and suspicious materials. Afterwards, we can quickly analyze and extract the data. Robots with artificial intelligence are more reliable than people, more hard-working and indefatigable. For the application of this scenario, our goal is to halve the number of security personnel in super-large parks.

② patrol inspection: in addition to security, the application of artificial intelligence can replace part of the daily inspection work. These robots with various sensing capabilities can transmit video, sound and smell in real time during the daily inspection. This information is transmitted to the background and the original collected data of the infrastructure for comparison and useful supplement. For the application of this scenario, our goal is to reduce the number of infrastructure inspectors by nearly 30%.

Supervision and control of ③ equipment: the realization of unattended requires a large number of automatic actuators, which is our focus to strengthen the area. Recently, I have visited many equipment manufacturers. For the automatic control of electric power, we can now use electric operation to control the switch of the cabinet, so that it can execute the switch of the cabinet according to our instructions. In drills or large-scale failures, the middle voltage side to the low voltage side, from the diesel generator to the cabinet can be executed automatically remotely.

To achieve unattended and artificial intelligence, there is a particularly relevant application scenario, which is the refrigeration automatic control scene. After the indirect refrigeration production, the air conditioning dewatering, from the original extremely complex control devices, simplified to dozens of control devices, greatly reducing the threshold of automatic control and intelligence. We can optimize our algorithm according to different regions and different sites, realize fine management, optimize the condition of refrigeration, and achieve energy-saving and stable operation and maintenance environment. This is not only a key point, but also the key to our exploration.

For equipment monitoring and management, I would like to list a case of battery health management. As we all know, all the equipment in the data center can be divided into three categories, one is chemical, one is electronic, the other is mechanical. Chemistry is the easiest and fastest to enter the stage of high failure, and once the failure of chemical equipment is the most fatal. The health management of battery equipment is the most typical management of "chemical" equipment. We collect the battery voltage, internal resistance, temperature three state quantities, 20 seconds once, a day is more than 10,000 data, a year is almost 5 million data. A piece of data is a point, a series of data is a line, and we are now monitoring nearly 100000 batteries, which is an extraordinarily large data facade. At this time, we can judge without human experience, can subvert the original causal relationship, and directly rely on this massive three-dimensional data to judge whether the battery is abnormal and whether it should be replaced. The platform can directly issue processing instructions to prevent problems before they occur.

"Huangdi Nei Jing" said: "therefore, the sage does not cure the disease, does not cure the disease, does not cure chaos." In the face of massive and complex data centers, the current way of human scientific artificial intelligence embodies this kind of prior management and eliminates the essence of ancient thought.

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.