中文English
Your current position: Home » News Center » Interview & Dialogue » Academician Wu Hequan’s exclusive interview: In-depth analysis of the new development trend and future focus of my country’s computing power industry
Academician Wu Hequan’s exclusive interview: In-depth analysis of the new development trend and future focus of my country’s computing power industry
Date of publication:2022-10-03     Reading times:299     字体:【

Wu Hequan, academician of the Chinese Academy of Engineering
Accelerate the transformation of computing power into productivity
Computing power, the ability to process data. Under the background of digital economy construction, computing power has become a new productivity after heat and electricity. 

From mobile phones and computers to automobiles, supercomputers, aerospace rockets, to artificial intelligence, data centers, and the Internet, computing power is everywhere. In 2021, the scale of China’s digital economy has reached 45 trillion yuan.

Computing power is not only an important component of information infrastructure, but also a platform for carrying data and computing operations, and it is also a comprehensive manifestation of computing, storage, and network capabilities in the digital economy era. Data centers and communication networks become critical public infrastructure.

Data show that by the end of 2021, my country will have more than 450 ultra-large and large data centers and more than 20 intelligent computing centers in use. In the distribution of global computing power, China ranks second, second only to the United States. Among them, China’s basic computing power and intelligent computing power accounted for 27% and 26% of the world respectively.

As of the end of June 2022, the total scale of data center racks in use in my country exceeds 5.9 million standard racks, the scale of servers is about 20 million, and the total scale of computing power exceeds 150 EFLOPS (EFLOPS refers to the number of billions of floating-point operations per second) . How to accelerate the transformation of computing power into productivity? How to consolidate the computing base for the development of the digital economy? “Outlook” News Weekly reporter interviewed Wu Hequan, an academician of the Chinese Academy of Engineering.

Wu Hequan believes that as a new productive force in the era of digital economy, computing power and its industry are developing rapidly. The next step is to improve the level of coordinated development of the national computing power network, so that data in the eastern region can be more conveniently transmitted to data centers in the central and western regions for computing and storage, and promote the transformation of computing power into greater productivity.

[Digital economy creates strong demand for computing power]

▌”Outlook”: What are the new trends in the development of my country’s computing power industry?

Wu Hequan: In recent years, with the acceleration of digital transformation and upgrading in various industries, the total amount of data in the whole society has exploded, driving the demand for data computing and analysis.

Especially in scientific computing and engineering computing, these professional scientific research fields generate a large amount of data and have extremely high requirements for computing power. Taking oil and gas exploration as an example, oil and gas exploration is simply to do CT on the surface. After a project is completed, it needs huge computing power to analyze massive amounts of data.

Artificial intelligence is also a large demand for computing power, especially “eating” computing power. Taking the application of deep neural network in urban traffic prediction as an example, we collect daily traffic big data and input it into deep neural network for modeling. The trained model can guide the traffic flow of vehicles in real time. The process of deep neural network modeling relies on extremely high computational complexity and storage space, and requires a lot of computing power support.

When Alphago defeated Li Shishi, it used 30 million chess games as training data. OpenAI, an artificial intelligence non-profit organization, released a report in 2018, stating that since 2012, the computing power used in AI training tasks has doubled every 3.43 months…

With the accelerated development of my country’s digital economy and the continuous deepening of digital transformation, the application scenarios of computing power continue to expand, and the empowerment effect of computing power is highlighted.

In general, the development of the computing power industry presents three trends: first, the forms of computing power are more diverse, and the demand for computing power resources is strong; Expand further.

[Hot and cold collaboration promotes “counting in the east and calculating in the west”]

▌”Outlook”: What is the current progress of my country’s computing power infrastructure construction?

Wu Hequan: From a macro point of view, every 1 yuan invested in computing power produces a return of 3 to 4 yuan. Taking the data of 2020 as an example, the scale of my country’s 2 trillion yuan computing power industry directly and indirectly drives economic output of 1.7 trillion yuan and 6.3 trillion yuan respectively. . Building computing power infrastructure has also become a strategic choice for many countries around the world, including my country.

In February of this year, four departments including the National Development and Reform Commission jointly issued a notice, agreeing to start the construction of a national hub node of a national integrated computing power network in 8 places including Beijing-Tianjin-Hebei, the Yangtze River Delta, and the Guangdong-Hong Kong-Macao Greater Bay Area, and planned 10 national hub nodes. The data center cluster, the “East Counting West Counting” project was officially launched. At present, my country’s computing power distribution is increasingly optimized, and a development pattern in which core regions such as the Beijing-Tianjin-Hebei region, the Yangtze River Delta, the Guangdong-Hong Kong-Macao Greater Bay Area, and Chengdu-Chongqing are coordinated and supplemented has basically been formed.

The positioning of the construction of the eight national integrated computing power national hub nodes is different. Guizhou, Inner Mongolia, Gansu, and Ningxia mainly build non-real-time computing power guarantee bases for the whole country, continuously improve the service quality and utilization efficiency of computing power, give full play to the advantages of energy and resources, consolidate basic guarantees such as the network, and undertake nationwide background processing, Non-real-time computing power requirements such as offline analysis and storage backup.

The four nodes of Beijing-Tianjin-Hebei, Yangtze River Delta, Guangdong-Hong Kong-Macao Greater Bay Area, and Chengdu-Chongqing mainly serve the implementation needs of major regional development strategies. Coordinated and sustainable development of land, energy, water, electricity and other resources, optimize the supply structure of data centers, and expand the space for computing power growth.

According to the plan, by the end of 2023, in terms of utilization rate, the average utilization rate of national data centers will strive to increase to more than 60%; in terms of computing power scale, the total computing power scale will exceed 200EFLOPS, and the proportion of high-performance computing power will reach 10%; in terms of energy efficiency, the newly built The PUE (the ratio of all the energy consumed by the data center to the energy consumed by IT equipment, that is, the energy use efficiency) of large-scale and above data centers is reduced to less than 1.3, and strives to be reduced to less than 1.25 in severe cold and cold regions; in terms of network delay, national hub nodes The data center end-to-end network one-way delay is in principle less than 20 milliseconds.

▌”Outlook”: How to improve the coordinated development level of the national computing power network, so that the data in the eastern region can be more efficiently transmitted to the central and western data centers for computing?

Wu Hequan: At present, most of the demand for computing power is concentrated in the eastern coastal areas where the digital economy is relatively developed. The data generated in the east can be divided into cold and hot data: hot data is mainly real-time calculation data, and cold data is non-real-time calculation data, such as data generated last year and the year before last. From an energy point of view, the natural resource endowment in the west is more suitable for building computing infrastructure, so cold data can be stored in the west.

In the promotion process of “East Counting and West Computing”, it is necessary to further strengthen the collaboration between the eastern and western computing power hubs and data centers.

In the west, a data center continues to work 24 hours a day, but it does not mean that it is always computing, and data storage is the norm. Therefore, in a sense, “counting from the east and calculating from the west” is actually “counting from the east and saving from the west”. However, due to the lack of a coordination mechanism, some computing power hubs in the west do not know how much cold data in the east will be stored in the west, whether to put it in Ningxia or Guizhou or other hubs, and need to do a good job in planning and coordination. Some eastern cities tend to keep cold data around for easy use. For this reason, it is necessary to establish the coordination of cross-domain data center capabilities of “East Data and Western Computing” to avoid mismatching of storage and computing resources.

In order to achieve the optimal allocation of computing power resources, in addition to the ongoing construction of the “East Counting West Computing” project, it is also necessary to build a computing network coordination system so that computing power can most economically adapt to the development needs of the digital economy.

One is the collaborative scheduling of computing and network resources. Usually, computing and network operators are different, and do not perceive each other’s needs and capabilities. It is necessary to adopt technologies such as the next-generation Internet protocol IPv6, which can improve my country’s Internet carrying capacity and service level, to perceive customer needs and optimize the selection of computing power nodes. ;

The second is collaboration among data centers. Including the configuration of master and backup in the same data center and remote disaster recovery, optical fiber transmission capacity between data centers and reasonable planning of dual routes, etc.;

The third is the coordination of computing power across administrative regions. For example, how to pair and coordinate data centers in the eastern and western regions;

Fourth, in terms of management, local governments should strengthen the optimization and coordination of “source, network, load and storage” including land, public energy, and power equipment;

The fifth is the coordination of energy layout and computing power layout. While meeting the demand for computing power in the east, it will drive the digital economy and industrial development in the west.

[Promoting enterprises to go to the cloud to achieve computing power catch-up]

▌”Outlook”: Where is the focus of my country’s future development of the computing power industry?

Wu Hequan: The first is to continue to promote the development of artificial intelligence computing power and consolidate the infrastructure of artificial intelligence computing power. In recent years, with the rapid development of my country’s artificial intelligence technology and continuous expansion of application scenarios, artificial intelligence computing power centers have also ushered in a rapid development momentum. Tianjin, Chengdu, and Shanghai have built local intelligent computing centers. In the distribution of computing power in 2021, the United States accounts for 35%, 15%, and 30% of the world in the three types of basic computing power, intelligent computing power, and supercomputing power, and China accounts for 27%, 26%, and 20% respectively. %, the United States is dominated by basic computing power, and China’s intelligent computing power exceeds that of the United States. In the future, we should continue to strengthen the construction of artificial intelligence computing power infrastructure to provide computing power support for artificial intelligence to promote the digital intelligence upgrade of the real economy.

Secondly, improve the digital ecology and stimulate the demand for data center applications. At present, the problem that my country’s enterprises dare not or do not want to transfer to the digital transformation still exists. The application of consumer digital scenarios needs to be deepened, the industry digital application scenarios need to be explored, and the digital business environment needs to be continuously optimized. Some local governments open up data and application scenarios and computing power resources, and encourage Internet platform companies to cooperate with small and medium-sized enterprises to provide complete cloud service solutions including consulting, design, manufacturing, verification, implementation, and operation, thereby lowering the threshold for using computing power , to solve the problem of enterprises migrating to the cloud. Transforming computing power into productivity can promote the pace of digital transformation of enterprises and create greater business value.

Third, the data center industry chain needs to be extended and optimized. As an asset-heavy industry, data centers have a slow return on investment. In the past, the development of data centers in the western region was mostly in the form of leasing digital space, leasing racks, hosting servers, and few value-added services. The two ends of the industrial chain application and market are outside, and the service links such as data collection, cleaning, mining, and analysis are underdeveloped, making it difficult to fully release the development potential of data elements.

In the future, the Western Data Center can extend the industrial chain around the central link and provide more value-added services: on the one hand, it will expand upstream to perform data preprocessing, including data labeling, cleaning, desensitization, etc.; on the other hand, it will extend downstream, including production Servers, edge computing, and engineering operation and maintenance services can also develop towards big data mining and applications.

Fourth, improve the level of green development. Under the background of dual carbon, energy saving and emission reduction is an unavoidable topic in the development process of the computing power industry. In the long run, the entire society needs a “digital base” composed of computing power and connections, as well as a “green base” constructed of clean energy and green technology. The two must develop in coordination.

At present, the energy consumption of data centers accounts for about 2% of the energy consumption of the whole society, and the energy consumption is still increasing, which requires us to use more advanced energy-saving technologies and introduce more green and clean energy. When building a data center, you need to pay attention to the PUE value, and at the same time, you need to work hard to reduce the energy consumption of IT equipment. It is worth mentioning that a low PUE value does not mean low carbon emissions. What we should pursue is to increase the proportion of renewable energy use with a low CUE (the ratio of the total CO2 emissions of data centers to the energy consumed by IT equipment).

In terms of computing power infrastructure construction, it is still necessary to make up for key technical shortcomings and talent shortages in chips, operating systems, etc., which will require long-term efforts.

Published in “Lookout” Issue 38, 2022

 
Previous:
                           Next: