Release Date：2021-11-30  Author：By Mao Lei  Click：

Research Background
As Internet technologies develop rapidly, computing carriers are more lightweight and dynamic, applications are decoupled into microservices, and network functions are cloudified, resulting in increasingly blurred boundary of computing and networking. With the business development of the intelligent society, massive raw data generated by various sensing terminals needs to be processed, and the demand of L4/L5 automated driving for computing will grow rapidly, which will promote the growth of computing infrastructure. Facing the future, ubiquitous computing resources, services, and networks need to be integrated efficiently so that the demand side can use computing resources in the same way as electricity.

China issued the "Guidance on Accelerating the Construction of Collaborative Innovation System of National Integrated Big Data Centers" in May 2021, which put forward "promoting service-based computing resources" for the first time, including building the integrated computing service system and optimizing the computing resource requirement structure. The concept of "East Data & West Computing" is proposed at the national level. To support national industry upgrade and digital transformation of the whole society, there is an urgent need to solve the problems of competition, coordination and win-win among computing, networking and applications, as well as maximizing resource efficiency.

Global standards organizations including ITU-T, IETF, BBF, ETSI and CCSA have made some technical proposals for application scenarios, requirements and reference architecture of computing force network, but they still need to be clearly converged.

Research Focus 
At present, computing and networking belong to two independent technical and operational domains. The networking serves as the center to schedule computing, storage, and network resources in a unified manner. Through the enhancement of the network layer, network devices participate in sensing and orchestrating computing resources, and the network can perceive computing power, thus providing new integrated computing services. 
The research focus of computing force network includes computing operation, computing-aware network, and computing power measurement, with the aim to achieve an end-to-end closed loop of computing from generation to scheduling, and finally to external transaction (Fig. 1). 

Computing Operation 
Computing operation means to integrate and transact computing resources, and manage and schedule ubiquitous computing devices and multi-level computing at the end, cloud and edge in a unified manner to form a distributed cloud and complete computing transactions for end users through the operation platform. To achieve intelligent scheduling of computing and networking, the operator's computing nodes in its data center, MEC and CO, as well as the third-party's computing nodes at the end, edge and cloud, are registered to the computing and networking orchestration management system through a trusted mechanism.
The blockchain technology is introduced and applied to computing access authentication and transaction to implement de-centralization computing sharing through the consortium blockchain. Ubiquitous computing nodes, through the permission, enter the link, shield illegal nodes and traffic, and protect data security. A trusted transaction is achieved through a smart contract. During a computing transaction, the operation transaction platform accounts out through the smart contract, and the data is input into blocks and recorded on the chain.
Computing-Aware Network 
Network equipment that participates in computing resource awareness and orchestration, is highly related to routing protocols. The network equipment needs to be reconstructed fundamentally. With the help of computing service routing and address-based routing, a native computing-aware network can be implemented. The computing-aware network can perceive global computing resources by dynamically releasing the resources of computing nodes such as edge DCs, regional DCs and central DCs to the network and coordinating with each other. The computing service awareness decision point, namely the computing service gateway, is introduced to resolve user service requests to computing resources and forward computing services based on the double constraints of computing and networking SLA. Network forwarding plane technologies such as SFC, ICN and SRv6 are also used to provide precise differentiated computing services through certain expansion.
—Computing route: The head node of the network needs to map computing service requirements of the application and encapsulate computing routes, and the routing layer needs to perceive the computing application and its requirements.
—Control plane awareness: Access control process such as IPoE/PPPoE is extended. Currently, only user access authentication is provided, which can be extended to perceive important user applications. BGP/IGP protocols are extended. The computing applications can be registered and authenticated in PE or access GW, and notified and synchronized through BGP/IGP.
—Forwarding plane awareness: The packet header encapsulates the application information and performs the corresponding forwarding according to the delivered policy.
Computing Power Measurement 
Computing measurement and identification is the basis for supporting computing transaction and operation, and there is no consensus or standard in the current industry. With the development of cloud computing, computing carriers are becoming more lightweight, and applications are lighter. In the age of computing and networking integration, the orchestratable computing granularity at the network layer should also be fine-grained and hierarchical. By quantifying the computing power, a mapping model from service to computing is established to meet the user's request to initiate computing service. The network resolves the computing service, calculates the mapping from computing service to specific computing resource, and thus implements efficient scheduling of ubiquitous computing resources. 
—Computing grading: Computing power is graded in fine-grained mode. The serviceable computing granularity includes service granularity (AI training, and video processing), function granularity (coding compression, and encryption) and atomic granularity (CPU, GPU, FPGA, and ASIC).
—Computing conversion: The conversion between computing should be provided at the level of functional granularity. Especially in the MEC scenario, when the computing resources of the same type are scarce and the services need to be expanded, the network gives priority to the conversion of computing within the MEC, such as the conversion from FPGA to CPU.

Conclusion
Network connections will be ubiquitous in the future, and there will be massive ubiquitous computing access networks. The ubiquity of computing is becoming true. The future network will be oriented to users and applications, and computing and networking need deep integration and collaborative development to maximize resource efficiency and optimize service experience. Through the computing force network, a new green, low-carbon and multi-level computing and networking infrastructure can be built.