The computing power network came into being, and the service model will change from "resource-based" to "task-based"

The 21st century is a period of vigorous development of information technology. Information has become a new resource for economic development, and computing power has become a new driving force for economic construction. Computing power refers to ubiquitous data processing and computing capabilities, which is a combination of multiple capabilities such as algorithms, communications, cloud computing, and speed. However, due to my country's vast territory, dense population density in the east and sparse population in the west, and natural resources in the east and west, this mismatch between resources and information enrichment has affected the development of my country's informatization construction.

In May 2021, the National Development and Reform Commission and other departments first proposed "East Data West Computing", and four departments jointly issued the "Implementation Plan for the Computing Hub of the National Integrated Big Data Center Collaborative Innovation System", proposing the intensive development of data centers in the east and leapfrog development of data centers in the west, realizing coordinated dispatch between the east and the west, launching the construction of 8 national hub nodes, conducting cross-regional computing dispatch, accelerating the implementation of the "East Data West Computing" project, and building a national computing network system. In July 2021, the Ministry of Industry and Information Technology issued the "Three-Year Action Plan for the Development of New Data Centers (2021-2023)", hoping to form a new data center pattern with a reasonable layout, advanced technology, green and low-carbon, and computing power scale that is compatible with the growth of the digital economy within three years.

This article introduces the international and domestic standard-setting status of computing power networks, the industry's insights into computing power networks, and the application directions of computing power networks.

The computing power network came into being

The computing power network is the operator's implementation of the national "East Data West Computing" strategic requirements. It takes computing as the center and the network as the foundation. It deeply integrates "network, cloud, data, intelligence, security, edge, end, and chain" (i.e., ABCDNETS) to provide integrated services. It realizes the transition from network-centric information exchange to computing power-centric information data processing. It is a key part of the intelligent, comprehensive new information infrastructure, as shown in Figure 1.

Figure 1 Computing network realizes ABCDNETS fusion innovation

According to the national strategy of "East Data West Computing" and the layout of multi-level data centers, a high-speed data center direct connection network will be established in the future to accelerate network interconnection and build a cloud-network integrated, network-integrated, and edge-access network with excellent performance. Specifically, it is necessary to promote resource linkage between multiple clouds, between clouds and data centers, and between clouds and networks to achieve cloud-network integration; build a data center direct connection network around the cluster, optimize the network structure, expand bandwidth, reduce data transfer delay, improve the level of cross-regional computing power scheduling, and achieve cloud interconnection; network performance and real-time performance are excellent, and the end-to-end one-way network delay of the data center is in principle within 10 milliseconds to achieve full cloud access.

In order to promote the "new infrastructure", integrate new elements, and stimulate new momentum, my country's three major operators have deployed computing power networks, further promoted the upgrading of information service capabilities and supply, and promoted the application of computing power networks in more fields such as national governance, social livelihood, upgrading and transformation of traditional industries, and domestic and international market expansion, so that computing power can become a social-level service that can be "accessed at one point and used immediately" like water and electricity. The computing power network will develop in three stages from the "ubiquitous collaboration" and "integration and unification" stages during the "14th Five-Year Plan" to the "integration and endogenous growth" in the medium and long term. Through the scientific and reasonable layout of data centers, the advantages of existing communication networks will be brought into play, and the maturity of computing power network technology, industry, ecology, and business models will be comprehensively promoted. Finally, a new service system of "connection + computing power + capability" focusing on 5G, computing power network, and smart middle platform will be built, and the formulation of computing power network standards, consensus, and application improvement will be actively promoted to truly realize that the network is everywhere, computing power is everywhere, and intelligence is everywhere.

At the 8th Network 5.0 Plenary Meeting in 2020, the China Academy of Information and Communications Technology, China Telecom, China Mobile, China Unicom, ZTE, Huawei, and the Chinese Academy of Sciences jointly established the Network 5.0 Innovation Alliance Computing Network Working Group, which mainly studies different routes and technologies in computing network integration, hoping to reach a consensus in the computing network industry and formulate relevant standards to promote industrial development. China Mobile has released the "Computing Network White Paper", and the China Telecom Research Institute has also launched research on the progress and trends of computing network integration. The computing network is developing in full swing.

The status of the formulation of computing power network standards

From a domestic perspective, before the country proposed the concept of "East Data West Computing", the three major domestic operators and equipment manufacturers had insight into the long-term development trend and direction of edge computing, and gradually reached a consensus that computing power networks will develop rapidly and become a growth point for the future digital economy. They placed the research on computing power networks at a strategic height and conducted in-depth exchanges and discussions on the basic concepts of computing power network technology. So far, the China Communications Standards Association has established six "industry standards" led by the three major operators, involving computing power network architecture, protocols, identification and transactions; the Network 5.0 Innovation Alliance Computing Network Working Group established at the 8th Network 5.0 Plenary Session in 2020 proposed an OverLay architecture based on DNS; the Computing Network Integration Working Group established by the China Communications Society also has two "group standards" projects. It can be seen that the domestic industry and academia have invested heavily in computing power networks and have achieved many research results. The promotion and maturity of some key "industry standards" will help promote the deployment and implementation of computing power networks.

However, the progress of computing network standards and research is slow internationally. Currently, the focus is mainly on the decoupling of cloud services and network connections by "big T", and there is insufficient motivation for the research and development of computing network integration. Therefore, it is possible that the domestic computing network will push back the international standard process after the vigorous development of computing network. Specifically, the COIN group established by IRTF mainly conducts research on network programming and new transport layer protocol architecture in computing network integration scenarios, but there has been no substantial progress; IETF proposed a distributed solution architecture, but it is mainly promoted by the three major domestic operators, ZTE, and Huawei; ITU conducts research on computing network architecture and scenarios, which is led by China Telecom and China Mobile.

The development and application of computing power network

As shown in Figure 2, the computing power network can be divided into the application layer, the computing power resource scheduling platform layer and the routing layer. The application layer completes the decomposition and application of computing power, the computing power resource scheduling platform layer completes the allocation of computing power, and the routing layer completes the forwarding of computing power.

Figure 2: Computing power network layer

Computing network architecture

The computing network architecture is divided into a centralized computing network architecture and a distributed computing network architecture, as shown in Figure 3. This architecture has the only computing network orchestration and scheduling center in the entire network, which has global resource information of computing power and network, and can coordinate the computing power and network of the entire network. First, the scheduling strategy generated by the computing network orchestration and scheduling center sends the routing forwarding table through the controller. At the same time, the controller needs to collect network information and send it upward to the computing network orchestrator. Secondly, the computing network orchestrator is responsible for collecting computing power information, and combines network information to perform computing network joint orchestration, and sends the orchestration strategy to the controller to complete business scheduling. Thirdly, in the centralized computing network architecture, the computing network orchestration center has a global resource view, so the impact of the equipment can be minimized. Finally, the computing network orchestration layer maintains the global static computing power and network topology information, and then synchronizes it to each local network node. Each local computing power routing node maintains the local computing power resources and network resource information in real time, so the local area computing power routing efficiency is high, but the subsequent equipment upgrade cost is also high.

Figure 3 Centralized computing power network architecture and distributed computing power network architecture

Since the computing network is still in the stage of continuous updating and iteration, it can be verified through centralized computing network deployment in the early stage, and distributed verification can be carried out in specific small-scale networking scenarios in the later stage. After the protocols, standards, and solutions of the computing network are fully mature, the large-scale deployment of the distributed computing network will be gradually completed.

Application of computing power network

For operators, computing power networks can help them transform their service models from providers of bandwidth and networks to providers of network computing capabilities, and seek new development directions by refining the conditions of each business in the operation network. For cloud vendors, computing power networks can help them achieve resource sharing, elastic on-demand mobilization, and save a lot of asset investment and operation and maintenance costs of distributed edge nodes. Therefore, computing power networks have extremely high commercial value.

However, the specific application of computing power network still needs to be explored, and the industry is still exploring. Through the construction of computing power network, the service model can be transformed from "resource-based" to "task-based". Through the integration and global scheduling of computing network resources, users can be provided with the most cost-effective integrated computing network service, which can improve the overall utilization rate of computing network resources for the company.

When the "resource-based" model is transformed into the "task-based" model, different modes of service can be provided according to the differentiated needs of users, including directly providing application API interfaces (APaaS capabilities) to users, which can be directly called when users apply them without having to pay attention to the distribution of computing resources. This is an all-inclusive model that charges according to the number of service calls; users provide application code packages, and the platform completes image packaging, distribution and deployment, and dynamic scheduling, and provides GPaaS capabilities to users. This is a semi-inclusive model that charges according to the number of application instances. The scheduling capability can also be directly opened to users to present the busy and idle status of resources, and users can independently complete application scheduling, that is, a capability opening model that charges according to resource supply specifications.

Technical Challenges Facing the Development of Hash Network

Since the computing power network is a brand-new network architecture, the concept and standard of "computing power" as the basic unit of measurement of the computing power network have not yet been reached in the industry, making it difficult to unify the management of computing power equipment, quantification of computing power, and computing power billing. This is a problem that the industry needs to solve. We should unify the measurement of computing power provided by different algorithms and resources, build a measurement system for users' actual computing power needs, and at the same time improve the resource performance indicators recognized by the industry to provide a unified computing power service model for the outside world.

Figure 4 Hierarchical computing resource measurement system

Based on the above problems, a hierarchical computing power resource measurement system should be gradually established. As shown in Figure 4, the global semantic computing power identification system includes computing power identification of the application layer, transport layer, and routing layer, which is the cornerstone of the computing power network architecture; and the hierarchical computing power resource measurement system is the basis for computing power transactions and operations. IaaS, PaaS, and cloud native resources and services all belong to the broad computing power and can be included in the unified orchestration and scheduling of the computing network platform. In the actual promotion stage, we can first focus on basic computing power, and gradually incorporate PaaS and cloud native basic services as the measurement and standardization process progresses to form a hierarchical computing power resource system.

The 14th Five-Year Plan emphasizes the need to strengthen "new infrastructure", build a national integrated big data center system, accelerate the construction of data center systems, and meet the application needs of various industries. A strong network and a digital China have become important strategies for social development at this stage. Operators' networks should also evolve towards an integrated computing network to serve the digital development of thousands of industries, achieve the goal of "computing power as a service, network as a platform", and empower and increase efficiency for all industries. The computing network is still in the stage of improvement, and there are still many problems to be solved. At the same time, operators should also strengthen academic and research cooperation, industry-research cooperation, tackle technical problems, and build a healthy and sustainable computing network integration ecosystem.