Calculate net you talk | why calculate forces network face a double challenge?
“The huge growth in computing power is challenged by both computing architecture and energy efficiency.”With the development of global digital economy, data has become a new factor of production, and computing power network, as a response to national strategy, to conform to industrial development and promote enterprise transformation, has brought new opportunities for the development of social digital intelligence and industrial transformation and upgrading.However, computing power network is not only a new network technology or a simple combination, but also a comprehensive and systematic project involving multiple product technologies in network, computing and platform fields, as well as standards and specifications, business scenarios, business models and cooperative ecology, which requires long-term development and evolution and joint efforts of the industrial chain.”‘ Connectivity + computing ‘, as a long-term strategic focus of Huawei, is highly compatible with the development direction of computing power network industry.”Recently, Xu Yongbo, head of Marketing department of Huawei China operator, pointed out in an interview with “Communication Industry Daily (net)” that Huawei will fully support the development of operators’ computing power network, be a core enabler of product technology, an important supporter of business and business model, and a synergist and promoter of industrial ecology.Based dual challenges facing the current work force, the digital economy has become an important part of our country’s economy, the “difference” plan and 2035 vision program for chapter 15 explicitly made the goal of creating new advantages of digital economy, to give full play to the huge amounts of data and application scenario for the advantage, promote the fusion depth of digital technology and the real economy,It enables the transformation and upgrading of traditional industries, gives birth to new industries, new forms and new models of business, and strengthens new engines of economic development. Among them, the development of computing power is paid more and more attention by the industry.Relevant data predict that by 2030, mankind will enter the YB data era, general computing power will increase by 10 times, artificial intelligence computing power will increase by 500 times, cloud services will account for 87% of enterprise application expenditure.The current huge growth of computing power is facing the dual challenges of computing power architecture and energy efficiency.First of all, with the failure of Moore’s Law, the growth rate between CPU, memory and I/O has been contradictory, and computing architecture is moving towards a new watershed. New computing architecture is urgently needed to break the existing contradiction.Secondly, it is the bottleneck of energy efficiency. The growth of computing power leads to the increase of heat consumption and the national green and low-carbon requirements have been contradictory. Under the premise of limited energy supply, better computing energy efficiency means more computing power development rights.In addition, as the digital transformation of the industry requires real-time and security of data processing, distributed deployment of computing power will become another major form of cloud computing.From the central cloud computing based on the Internet, to the edge computing with agile intelligence guaranteed by SLA network, to the future evolution into the linkage of cloud-edge collaborative capability and end computing.Undoubtedly, it is necessary to realize cross-computing node collaboration through computing network into network, and achieve a breakthrough on the existing computing architecture through distributed parallel computing capability across nodes.At the same time, the power network can solve the problems of power shortage and double carbon by using the high-speed and efficient optical fiber network.As we all know, computing power network is “a new information infrastructure that can allocate and flexibly schedule computing resources, storage resources and network resources among cloud, network and edge according to business needs”. Just as the development of electric power cannot be separated from the power grid, the development of computing power cannot be separated from “computing power network”.It is widely believed in the industry that the overall architecture of computing power network consists of “computing, network and brain”. Xu Yongbo gave a detailed description of the key technologies and development goals.First of all, the network is the infrastructure base supporting high-quality connection, fast access, flexible scheduling and arrangement of computing power. It is mainly composed of two parts, including the unified intelligent IPv6+ network and the ultra-wide and flexible all-optical base network.These capabilities are enabled by IPv6+ technologies, including well-known FlexE hard slicing, SRv6, on-stream detection, security, and AI-related capabilities.For users or business scenarios, the all-optical base network should not only support the flexible connection requirements of upper-layer IP, but also meet the requirements of scenarios with large bandwidth, long distance interconnection of computing power nodes and users with high-quality directly connected computing power nodes. It is also the basic guarantee for maximizing the value of computing power network and giving full play to the computing power network capability.Secondly, the goal of the computing field is to build a green, efficient and safe computing power infrastructure, unified computing power scheduling, and create a network of computing power.In the field of computing, efforts should be made to develop diversified computing forces, build an independent and controllable industrial ecology, innovate computing architecture, and break through the bottleneck of hardware based computing forces. In addition, computing forces should be unified and coordinated to realize the integration of computing forces, and industrial standards and specifications such as computing force identification and computing force quantity should be established.Xu yongbo specifically pointed out that AI can be the first to try, as the priority layout of the diversity of computing power network.In terms of green, it can achieve the ultimate PUE by innovating PoD hardware of the whole cabinet, accurately grading and efficient liquid cooling, optimizing DC infrastructure + linkage of the whole stack, and realizing the optimal cost of the whole network through cross-region scheduling of computing power network based on SLA requirements.Finally, the computer network brain is the key to support the computer network service, it realizes the unified arrangement of computer network and flexible resource scheduling.The network brain connects to the network enabling layer and the resource management and control system.Computer network brain includes important functions such as perception, arrangement and scheduling.Xu Yongbo puts forward some suggestions for the brain architecture of computing network, such as logically introducing the enabling layer of computing network to build the identification, measurement and management execution capability for computing power resources/network resources/data resources;For the converged scheduling strategy, more dimensions (energy consumption, tariff, etc.) besides network number resources should be considered.It is true that a mature technical foundation is the key to the promotion of business scenarios. However, without certainty scenario-oriented technology, it is more like a plant without “soil”, and the clarity of business scenarios is undoubtedly more targeted for the iterative evolution of technology.In this regard, Xu Yongbo proposed five types of business scenarios.One is the AI super-large model training, east and west calculation/storage scenarios.The data provided by the user is transferred to the data center or the western node, and the computing force network is used to realize the scheduling of the recalculation force.This requires computing power to achieve cross-domain clustering capability, and the network can provide super long and large bandwidth capability.The second is the industry number intelligent scene such as industrial control, machine vision.Industrial user data requires real-time processing, which requires the application of closed loop at the edge computing nodes or terminals in the near field.Computing power network should support real-time edge computing power service and user-edge before deterministic network.The third is the user game/entertainment scenario, such as cloud games, metaverse.Data is distributed from DC to edge and further transmitted to users. The edge side provides rendering, storage and other services, but also provides user-edge-cloud deterministic network experience.Fourth, we will optimize large ISP services.Business plan, such as game company need to rent frame, the installation location, hardware and software, monitoring state of the cluster, load balancing design, data synchronization, fault tolerance and the user experience optimization scheme and so on a series of work, if there is a power network, can make the task to calculate net, customers need to focus only on the business logic itself.Fifth, data trading.Through computing power network, the whole life cycle management of data can be realized, data + computing power service can be provided to data users, and data intermediary service can be realized.Xu yongbo pointed out that many scenarios of computing power network are still in the exploratory stage and need to be explored, studied and realized together with operators and partners.There is no doubt that with the transformation of network development from information exchange to information data processing, computing power will become the core and productivity of information technology development. To provide digital intelligence services requires operators to provide new information infrastructure with computing power as the carrier and multiple elements.However, at the macro level, there are still many constraints in the four aspects of computing power network. At the technical level, the identification measurement system of computing power network, the standard and model of computing power network brain are still in the research stage.The actual business requirements and scenarios of “network adjustment” at the business level are not clear.At the commercial level, the calculation and settlement between operators and customers and partners, as well as the construction of new business and cooperation models need to be explored;At present, power network at the ecological level is mainly promoted by operators and equipment suppliers, which needs the attention and input of more industrial chain partners such as OTT, industrial ISV/IHV/SI, and end customers.In this regard, Xu Yongbo put forward four suggestions.First, we will step up research and formulation of technical standards and specifications and accelerate pilot exploration of relevant new technologies.For example, jointly define the standard specification of computing force identification and computing force quantity with operators, and the standard specification of computing and network cooperation.Support the pilot deployment of new technologies such as application awareness (APN6) and computational routing (CFN).Second, it is recommended to speed up the deployment of basic capabilities around the target architecture, and plan and implement them appropriately in advance based on business drive.For example, the unified load bearing and flexible route selection of SRv6 are promoted, the slicing network of E2E is constructed, and the second plane of diversified AI computing power is created.Third, business around the scene to explore, promote commercial pilot and closed loop.For example, ToB/toC/toH joint investigation of all kinds of customers and scene needs, design and improve network related businesses based on machine vision, cloud mobile phone, cloud games and so on.Four is to strengthen the joint innovation on the industry, advice and support the operators polymerization equipment business, many resources, such as industry partners and customers around the standard specification, key technology, key scenes, through the joint innovation center, joint innovation projects and project cooperation innovation explore related to calculate power network in accelerating technology and business.Xu Yongbo, Head of Marketing department of Huawei China Region, has more than 24 years of experience in market insight, Marketing and solution development in the telecommunications industry. He joined Huawei in 2005 and has 10 years of resident experience in Africa, Middle East and Asia Pacific.He served successively as Marketing director of MTN System Department of Huawei, MKT Director of Southeast Asia Region Department of Huawei, and director of consulting Office of BG, huawei’s operator.Huawei’s “Ten Masters of Blue Blood” and “God rewards those who help themselves” winners.Behind the opening ceremony of the Winter Olympics: How strong is this “big connection”?What is behind the “clearest” Winter Olympics in history?