Computing network: The "symphony orchestra" in computing resources

Source: People's Liberation Army Daily, China Military Trumpet

Author: Jiang Ran, Yao Changsong, Yang Xiangrui

Conceptual diagram of a computing power network.

Not long ago, the International Organization of the World Internet Conference held a special forum in Barcelona, Spain, with the theme of "Computing Network: Intelligent Network Empowers a Smart World".

In fact, many aspects of our lives are closely related to computing power: when you sit in front of the computer and immerse yourself in the virtual world of a large game, it is the computing power behind the smooth gaming experience that is silently supported by the computing power, and when you watch a high-definition movie online, computing power helps you feel the high-definition picture quality and powerful sound effects, so that you don't miss a wonderful moment......

In today's era, the process of digital transformation and upgrading of the economy and society is accelerating, and new technologies and applications such as 5G, big data, and artificial intelligence have an increasing demand for computing power, and the supporting role of computing power in economic and social development is becoming more and more prominent. Today, the traditional computing mode can no longer meet the needs of a large number of computing resources, and the "computing power network" came into being and is developing rapidly.

So, what is a "computing power network"? What are the advantages of "computing power network"? What are the future development prospects? Please see the interpretation of this issue.

Make the "cloud" more orderly and the "cloud" easier to land

Computing power refers to the amount of information and data that the device can process per second according to the change of the internal state, which is simply the computing power.

As early as the end of the Eastern Han Dynasty, a kind of computing power based on human power appeared in Chinese society - abacus. With the abacus, people's calculation speed was significantly improved, and most of the computing problems in the society at that time were solved.

In the middle of the 17th century, the French mathematician Pascal designed a machine that could automatically complete calculations through mechanical motion, which was the world's first mechanical computer. In the mid-40s of the 20th century, the Department of Electrical Engineering of the University of Pennsylvania developed an "electronic numerical integration and computer" for the calculation of the ballistic trajectory of artillery shells for a ballistic research laboratory of the U.S. Army Ordnance Department.

At the same time, the development of integrated circuits and the proposal of the von Neumann architecture brought the development of the computer industry into the "fast lane", and the computer became the most strategically valuable industry in the world at that time. Gordon Moore, one of the founders of Intel, once put forward "Moore's Law" to summarize the speed of information technology progress at that time: when the price remains the same, the number of components that can be accommodated on an integrated circuit will double every 18~24 months on average, and the performance will also double.

As technology advances, the growth trend described by Moore's Law is becoming increasingly physical limited, and the physical limits of chip process technology are slowing down the increase in the number of transistors on integrated circuits. The demand for computer performance is getting higher and higher, and the technological progress of chip manufacturing alone can no longer meet the demand.

In this context, in 2006, Eric Schmidt, the former CEO of Google, first proposed the concept of cloud computing. Cloud computing integrates servers, storage and other resources on the network into a "cloud" through the Internet, decomposes the huge data computing and processing programs into countless small programs, and hands them over to the computing resource sharing pool for searching, calculation and analysis, and then transmits the processing results back to users. To put it simply, cloud computing is about breaking down tasks into different parts, computing them on different servers, and then merging them.

The computing power network is an upgraded and application version of cloud computing. Generally speaking, the computing power network is more three-dimensional and ubiquitous, and in the form of computing power clusters, it realizes the collaboration, intensification, and inclusiveness of computing power. From cloud computing to computing power network, "cloud services" have become more high-quality and accurate, and "cloud applications" have more and richer application scenarios.

Make the "cloud" more orderly and the "cloud" easier to land. The emergence and development of computing power networks provides better computing power support for the current hot artificial intelligence large models. In the face of the low utilization rate of computing power on the whole network and the diversified computing power needs of different users, the computing power network has become a new and more efficient solution.

As the core foundation of the new digital infrastructure, the computing power network is like a huge "symphony orchestra", in which massive computing resources are involved, and together they converge into an organic whole, using data as the musical score to play a beautiful "computing power movement".

"Bigger and stronger" and "flexible and precise" go hand in hand

The computing network is actually composed of three parts, in addition to "computing" and "network", the "brain" is also introduced. "Computing" is composed of computing resources such as CPUs and GPUs, which are used to produce computing power; "network" refers to network devices such as routers and switches, which are used to connect computing power; and "brain" is a control device, just like the "military division" of the entire computing network, which uniformly orchestrates, schedules, manages, and operates and maintains computing power networks and resources.

The computing network is like a "supercomputer", which first gathers the computing power of the whole network, and then uses the "brain" to reasonably distribute the data to each computing unit of the "supercomputer".

Due to the convergence of computing resources across the network, the most significant advantage of the computing network is that it has large-scale computing power and supports large-scale computing tasks. At the same time, the computing network also has excellent elasticity and scalability, and can flexibly respond to various complex scenarios and needs.

On the one hand, the computing network can flexibly provide users with the most suitable computing resources according to their actual needs: if it is a small data computing task, the computing network can provide small-scale computing power, and if it is used to train large-scale deep learning models, the computing network can also quickly mobilize resources and provide corresponding computing power. There is a vivid analogy that says: "The computing power network is like a 'universal faucet', users do not need to worry about other limitations, and only need to adjust the 'switch' according to their own needs, and they can realize computing services as they go." ”

For example, in 2023, China Unicom will independently develop a computing network brain management and control system to provide customers with ultra-low latency, ubiquitous access, and intelligent and agile computing power networks. It is reported that the computing power network has assisted Huashan Hospital in successfully developing the first surgical navigation medical algorithm model, which is used to simulate the surgical operation process, assist doctors in formulating surgical strategies, assist in analyzing and mining massive medical data, and discover trends and laws in the medical field, etc.

On the other hand, the computing power network allows the network to seamlessly add new nodes according to business needs, and improve computing power through flexible expansion. This dynamic expansion capability ensures the stability of the system, so that it can respond quickly in the face of burst traffic or large-scale data processing tasks, especially in cloud computing, big data, artificial intelligence and other fields, its wide range of applications is more eye-catching.

For example, Google's distributed file systems such as GFS and HDFS, as well as computing frameworks such as MapReduce and Spark, can be easily scaled to support the storage, access, and processing of large-scale data, ensuring the sustainability and resilience of its development.

Compared with cloud computing and edge computing, computing power networks combine the advantages of both, making data processing more efficient and flexible. Although the cloud computing model has a lot of computing power, uploading all data to the cloud for processing increases the time delay of data transmission, while edge computing effectively reduces the latency by moving data processing to the edge of the network, but its computing power is relatively limited. The computing power network draws on the essence of both, and through intelligent task distribution and resource scheduling, it can quickly determine the type of task and allocate resources reasonably. This precise task processing strategy not only improves computing efficiency, but also ensures the real-time and security of data.

"Bigger and stronger" and "flexible and precise" go hand in hand. The accurate and efficient computing execution of computing power networks in response to large-scale and complex computing tasks has attracted people from all walks of life who want to quickly obtain reliable computing power.

The integration of "computing", "network" and "brain" has broad prospects for military applications

With the acceleration of military modernization, informationization and intelligence have increasingly become important characteristics of modern warfare. As a data resource scheduling platform and future information infrastructure, the computing power network will also gradually be integrated into information-based joint operations, the construction of military organizational forms and the development of military scientific research, and play an important role in "empowerment".

Training determines combat effectiveness. The use of "science and technology +" to conduct simulated training can not only reduce the risk of resource consumption, equipment wear and tear and casualties, but also create a confrontational environment similar to actual combat without time and space restrictions. At present, the VR virtual simulation technology commonly used, the closer the simulated environment is to the actual combat, the more it needs the support of huge data. Using the huge computing power provided by the computing power network, it can realize fine scene modeling, real physical simulation, and realistic light and shadow effects of the simulated environment, provide military personnel with an immersive virtual training experience, and help commanders and fighters become familiar with various combat scenarios.

For example, in the 90s of the last century, the US Army invested nearly $1 billion to develop a networked combat simulation training system, which is the largest distributed interactive simulation system of the US Army so far. Now, with the blessing of the computing power network and the distributed interactive simulation technology, the system can establish a more refined and realistic virtual combat environment, support military training more intelligently and efficiently, and effectively improve the actual combat ability of the troops and the ability to deal with complex situations.

Today's massive amounts of battlefield data require faster information storage, processing, and analysis capabilities. The computing power network can give full play to its own advantages, quickly process massive intelligence information, and improve combat efficiency and accuracy by accelerating information fusion, assisting decision-making and operational command. For example, in the field of radar detection, a large amount of signal data generated by the radar system requires complex signal processing and target recognition, and the powerful computing power of the computing power network can help the system identify and track the target more accurately.

In addition, the wide application of large models further highlights the importance of computing power networks. Computing power networks can be trained and inferred with the help of complex machine learning algorithms and artificial intelligence models to better extract useful information from complex data, thereby speeding up the command and decision-making process.

For example, Scale Donovan, a large language model from Scale AI, can help warfighters, analysts, and decision-makers improve their understanding of the battlefield situation and the speed of action. In May 2023, the U.S. Army placed the Scale Donovan system on its 18th Airborne Division's encrypted network, primarily to help the division's units make decisions.

The key to the computing network is "interconnection". From a global perspective, computing power networks are still facing development challenges such as bottlenecks in resource scheduling technology and immature industrial ecology, and their military applications are also in their infancy. Despite this, the development trend of computing power network is beyond doubt that it integrates "computing", "network" and "brain". With the continuous progress of technology and the continuous expansion of applications, the computing power network, as the key "infrastructure" of the information age, will affect and even determine the digital, networked, and intelligent development process of various fields, drive the transformation and upgrading of various industries, and bring huge development opportunities while giving birth to applications and changes beyond people's imagination.

(Produced by the People's Liberation Army Daily, China Military Horn)