A new topic of computing power, let’s talk about the new people’s livelihood of computing power

01 2023 Hello!

This is a particularly meaningful beginning of spring. The whole people are overcast, everything is recovering, and the first open battle after the new crown has been won. From the beginning of the epidemic, the continuous struggle against the epidemic, to the victory of this battle, there is an invisible and intangible weapon that has been supporting us, and that is computing power.

What exactly is computing power? What kind of new livelihood can it bring us? Don't worry, let's start with counting the background and expand gradually.

02 What is East and West?

The national strategic project of counting from east to west has been launched for a long time. I believe everyone has heard of it. So what exactly is it? "Number" refers to data, and "calculation" refers to computing power, that is, the ability to process data. "East counting and west computing" is a new type of computing power network system that integrates data centers, cloud computing, and big data. The demand is guided to the west in an orderly manner, the layout of the data center construction is optimized, and the coordination between the east and the west is promoted.

"Numbering from the east to calculating from the west" has the same national strategic significance as the "South-to-North Water Diversion", "West-to-East Electricity Transmission", and "West-to-East Gas Transmission".

03 Similarities and differences between computing power network and State Grid

 "Digital from the East and Computation from the West" promotes collaboration between the East and the West through a new type of computing power network system, and is a national strategic project like "Transmitting Electricity from West to East". So what are the similarities and differences between the computing power network and the State Grid?

Four similarities:

• are pipeline resources

• Pipelines are connected to resources. For example, the national grid is connected to power plants, and the computing power network is connected to the computer resource pool of the computing center.

• resource provider

• are shared social resources

Five differences:

• The flow medium is different : what really flows through the grid pipes is electricity. What flows in the pipeline of the computing power network is not computing power, it is a byte called "byte", which is not computing power. Strictly speaking, the computing power network is not a pipeline resource of computing power.

• The direction of flow is different : the resource flow of the power network is one-way, and electricity flows from the power plant to our home, and it will not flow back. The computing power network is bidirectional. Our bytes can flow from the computing center to our client, and our client can also send and upload our files and data, which is equivalent to byte flow to the computing center.

• Different standards : grid output is standard, the frequency of electricity is 60 Hz, 220 volts. It is still difficult to form a standard for computing power, because the applications we use are diverse and the resources are also very complex, but with the advancement of scientific research, the future can be expected.

• Different status : State Grid is almost a monopoly in power supply. It is difficult for the computing power network to be like the State Grid, which can let the major computing centers just give them resources and then let them sell them, because they can also operate the major computing centers, so it is difficult for the computing power network to form a kind of Monopoly.

• The flowability is different : the power of the power network flows out, it flows to the electrical equipment through the power, and is consumed by the electrical equipment, that is, it flows out of a resource pool. But the computing power network is just the opposite. The computing power cannot flow out. Its computing power is in the computing center. If we want to use the computing power, we need to move or send our requests from the client to the computing center for use. This computing power.

Taking the power grid as an example, we have the concept of west-to-east electricity transmission. For example, when the electricity in the west is rich, it can be sent to the east, but we cannot send the computing power from the west to the east. On the contrary, we call it counting from the east to counting from the west. We move our data, the data in our east, and the requests and needs of more developed areas to the west, and use the computing power in the west. This is a very big difference.

04 The application of digital intelligence to count east and west

So, after understanding the concepts and differences, what typical application scenarios does it have? What are the characteristics of each?

Four typical application scenarios:

• East Digital West Training : Build a multi-modal pre-training model of speech, semantic meaning, and graphic text in the west, realize the unified representation of the three modes, apply it to scenes such as the Eastern Metaverse, and realize "sound with pictures" and "sound with pictures" through cloud-edge collaboration Diagram of sound generation". The feature is that both computing power and data are in the west, and the results of big data training are returned to the east.

• Storage in the east and west : The state stipulates that medical image data must be stored for 15-30 years, and the PACS image archive data of medical customers can be stored in the low-cost western data center. The characteristic is that warm and cold data are stored in the western resource pool.

• East View West Rendering : For customers such as film and television post-production, animation production, etc., it provides channels for uploading original materials to the east and transferring them to the west, and allocates rendering computing power in the west to complete the rendering task with cost priority, and returns the results to the east. The characteristic is that the massive CPU and GPU computing power required for rendering is hosted in the western resource pool.

• Side-by-side training : Industry, transportation and other fields require end-to-end delay applications within 10ms, and deploy them to 5G edge nodes. At the same time, upload data to the center for training and obtain a new model. The feature is that high real-time requirements require computing power to dispatch on-site or near-field resource pools.

Scenario application characteristics can be summarized into two points:

• Delay -insensitive computing tasks and infrequently accessed warm and cold data are dispatched to the low-carbon, low-cost west for storage and computing, and only the computing results are returned to the east.

• Delay-sensitive computing tasks are dispatched to the on-site or near-field to meet the real-time requirements of production.

05 The scene around me

After understanding the typical scenarios, let's go closer to the life scenarios we feel personally to see how our computing power weapons can protect our lives.

Computing power and virus speed

In December 2019, after the outbreak of the epidemic, from the country to the civilians, Qixin worked together to fight against the virus, saving patients, researching vaccines, and preventing the spread.

Nucleic acid detection has become the most important task, but at the same time it is unimaginably difficult. A suspected case needs to compare 22 million fragments. How can we use technology and viruses to "rob" precious time?

In March 2020, more than 120,000 cases were diagnosed worldwide, and the computing power of scientific research institutions was limited. Alibaba Cloud opened the gene computing service to the world for free, and it only took 60 seconds to complete the genetic comparison of any virus, including the new coronavirus.

It is this powerful computing power service that compares the speed with the virus, detects it efficiently, promotes timely isolation, prevents spread, and protects our lives.

Computing Power Supports Epidemic Management

During the epidemic, the social norm is home isolation, telecommuting...

The office model that is separated from the internal network environment of the enterprise poses a challenge to the digital level of the enterprise. In the new "boundaryless" enterprise network structure, it is as large as the operation of the enterprise data center, as small as the interconnection of a digital office equipment, a remote Behind every common office operation is the support of computing power.

Facing the surge in the cloud service market, the cloud platform hastened to expand its servers.

DingTalk added 10,000 new servers on the first day of the outbreak in 2020; 4 days later, another 10,000 were added; on February 17, DingTalk had expanded the capacity of more than 100,000 cloud servers on Alibaba Cloud.

The Tencent conference expanded to more than 100,000 cloud hosts in 8 days.

Kingsoft Documents Teleconferencing Server expanded by 3 times.

During the outbreak, the Zoom video conferencing app was downloaded 8.5 times more than usual in China.

During the outbreak of Ketianyun, the number of active users increased by 10 times, and the number of consultations increased by 20 times...

Major manufacturers have invested in cloud computing. On April 20, 2020, Alibaba Cloud announced that it will invest another 200 billion in cloud computing in the next three years, which means 5.6 billion a month. This reminds us of the movie "The Richest Man in Xihong City" Liwangduoyu challenged the story of spending 1 billion yuan a month, while Alibaba Cloud challenged spending 5.6 billion yuan a month. We think the important reason why Alibaba Cloud invests so much in cloud computing is the high level of the industry. Growth and market prospects are great.

In 2019, Alibaba Cloud’s revenue was US$5.2 billion, a year-on-year increase of 64%, making it the third largest cloud computing company in the world. Not only Alibaba Cloud, but now the performance of major cloud giants is growing rapidly. Coupled with the epidemic situation, hospitals have to go to the cloud, educational institutions have to go to the cloud, and online offices have to go to the cloud. On February 5, 2020, the number of downloads of DingTalk jumped to the top of the Apple App Store. The mobile terminal market rushed to the first place for the first time.

06 Imagined scene

The new life of digital intelligence empowered by computing power is far more than related to the epidemic. Next, let’s imagine an extreme car experience scenario that is more people’s livelihood, and see how we can use technology to create a better future.

This scene was actually mentioned by Sun Ninghui, an academician of the Chinese Academy of Sciences, at the Yunqi Conference not long ago.

The development of computing power services is analogous to upgrading the experience of buying a car, which can be divided into three levels:

• Buying a car: more complicated . You need to understand many technical details of the car, listen to a lot of sales, and only then can you choose a favorite one with a lot of calculations, and it costs hundreds of thousands. The underground garage is idle from time to time, and it is depreciated every year. At this level, consumers of computing power can only obtain computing resources through physical hosts. They need to know a lot of professional parameters when purchasing, which is difficult to select, requires maintenance after purchase, and may be idle from time to time, wasting resources.

• Renting a car: Relatively easy . As long as you care about the model and general functional parameters, you don't need to know many technical details. At this level, computing power consumers still need to care about the supply methods of various computing resources.

• Car use: minimalist experience . The ultimate service of unmanned driving, as long as you care about where you are going, when you need to arrive, whether you want to watch a movie in the car, have a meeting, or just lie in the car and sleep, it is an experience of using a car. At this level, consumers of computing power do not need to care about computing resources, they only need to put forward their own needs, and computing power network realizes the automatic allocation and management of computing resources according to the description of the needs of computing power consumers (such as destination, time requirements, and car experience requirements) .

The development of computing power services is inseparable from the research and breakthrough of new computing power network technologies.

To achieve the status of car rental, it is necessary to realize the separation of the website and the network, the website refers to the computing power provider, and the network refers to the computing power consumer; it needs to support the key technologies of four types of ecological roles. Suppliers need to achieve standardized intervention and single measurement; virtual operators need to achieve flexible orchestration; value-added service providers and consumers need to implement better payment models and convenient use.

In order to achieve the state of car use, it is necessary to realize the transformation of rental into use, which requires key technologies such as computing power network native programs and computing power network cloud functions. To put it another way, it can be used as you need, it is very flexible, and it is easy to change from small to large. You can directly describe your needs, and the computing power network can automatically allocate and manage computing resources according to your needs, or it can change with your needs (for example, I want to watch a movie in the first 2 hours of driving, and I want to sleep in the next 1 hour of driving) Allocate resources dynamically.

Our current cloud computing is basically still a car rental model, and we can’t get to the car model for the time being, but with the current research and implementation of new computing power network technologies, the future is worth looking forward to.

07 Future speculation

As the computing power network continues to accelerate and heat up, actively introduce cloud-native, serverless computing, heterogeneous computing, computing power offloading and other technologies, explore new directions such as computing power native, storage and computing integration, continue to enhance computing power, and release computing power Value; computing power procurement system, computing power pricing system, computing power trading model, pan computing power ecological cooperation system, etc. will all produce new changes. In the future, more new computing power ecological scenarios will emerge, bringing more intelligent and senseless service experience.

In addition to scenarios such as life, industry, and society, computing power also efficiently empowers the research and development of digital intelligence applications, because the infrastructure is jointly solved by multiple computing power resources, with high resource utilization, flexible resource scheduling, and agile resource expansion and contraction. serverless computing services. It can be used as you need, it is very elastic, and it is easy to change from small to large. You can program to specify what kind of computing power experience you need, what kind of charging standard you need, and what kind of network you need. Programming, in this case, the user experience will be better, and it will be more in line with our on-demand use. We can pay only for execution, develop in any language, simplify the development cycle, simplify the development and operation cycle, cost-effective performance, usage fully visible etc.

With the evolution of heterogeneous interconnection, cloud-edge/edge-edge collaboration, computing power and network collaboration, and social computing power grid integration, a multi-integrated computing power service network revolution will eventually usher in.