Throughout this year’s Two Sessions the phrases ‘computing network’ and ‘channel computing resources from east to west’ kept circulating. China has treated computing capacity as a priority for years, the east-west transfer project has formally started, and the country is building a computing network oriented around data flows. Opening the data artery and scheduling capacity as one coordinated system is a job the information and communications industry, as the main force behind the network, has taken on. Capacity begets capability: faster build-out releases the innovative value of data as a factor of production, supports network quality, new industry applications and ubiquitous device connectivity, and puts new momentum into the digital economy.
Treating capacity as infrastructure
How do you schedule capacity as one system? By turning it into infrastructure. Digitisation generates data at a volume that demands a converged computing architecture able to offer varied and differentiated services, and that is what a computing network is for. Under the new-infrastructure agenda, computing has been raised to the level of water, electricity and gas. ‘Build infrastructure moderately ahead of demand’ is a lesson Chinese transport planning has already proved; computing, as the new productive force of the digital economy, needs the same treatment so that it can carry industrial upgrading. The network is the artery linking users, data and services, and its convergence with computing is now the trend: only by hardening the digital infrastructure and giving the network the means to manage and allocate capacity do you get a network that is ubiquitous, layered and available on demand, and with it the room to rebuild the network around data.
The strategic case is not in doubt. But as new network services multiply, demand surges and computing requirements grow more complex, deep convergence of cloud, edge and device becomes both a requirement and the only sensible choice. A computing network spans heterogeneous hardware and chips, access and transport networks, data centres and cloud, plus big data, AI and blockchain; it touches container, microservice, cloud-native and DevOps practice. Designing the whole of it is a systems engineering problem on a large scale. As Zheng Weimin puts it, the computing network is both a near-to-medium-term plan and a longer-term direction for innovation, and it has to be built up steadily.
In aggregate China’s infrastructure is among the largest in the world, but supply is structurally short — regional imbalances, a low share of distributed capacity — so capacity per head remains low. What is needed is a digital infrastructure that integrates cloud, edge and device capacity on a high-speed, mobile, secure and ubiquitous network base, and delivers sensing, transport, storage and computation as one service, so that different workloads get the coverage, capacity and latency they need.
Today the computing industry and the network industry are still largely separate. Computing networks are at an early stage on standards and architecture, and barriers remain: data is not shared well, resource interfaces are not unified. Telecom operators and capacity providers need to work through the fundamentals — how capacity is measured, how it is traded, how computing and network are scheduled together — and to set standards for them. They also need a workable scheme for opening capacity up and sharing it, one that solves the trust problem in trades between different operators, so that data and information stay safe as the two converge.
The foundation matters more than anything built on it. Assembling the national grid also means developing systems for managing capacity pools, for measuring and brokering capacity, and for forming the core technology, base software and key systems that infrastructure requires. These are both the precondition for a computing network and the ground on which the digital economy is built. Testbeds should run in parallel, so that each critical technology can move from incubation to maturity and form a technical system that can be replicated.
Scheduling services inside the network
Getting the national grid right depends on timing the moves. The deep convergence of cloud, of the applications and services that run on it, and of the carrying network is the foundation of digital transformation and the core of the new information infrastructure. A computing network sits on ubiquitous connectivity and schedules network, storage and compute as one, matching capacity to the services inside the network on demand; cloud and network are the two bases, and they reinforce each other.
In cloud technology, orchestrating virtual resources is the baseline; container and capacity orchestration is the direction. Unlike automation, which typically targets a single task, orchestration handles an end-to-end process — managing the related services, guaranteeing availability, handling post-deployment work and recovery — so that a user can deploy and start using a service on a server or on any cloud platform. Capacity inside a computing network is ubiquitous and heterogeneous: CPU units on x86 servers, GPUs with strong parallel throughput, FPGAs. Because they run different algorithms on different data types, a mapping between unified capacity identifiers and network identifiers has to be established before heterogeneous resources can be scheduled as one system, and that mapping is the technical base for trading capacity.
Rising demand pulls the industry forward. Meeting the need for flexible hosting and intelligent scheduling calls for a network built around capacity, combining SRv6 (segment routing over the IPv6 data plane) with application awareness, so that compute and network can be supplied flexibly and quickly. A national project like the east-west transfer has to match capacity to workload by managing how data moves, according to supply and demand inside the network. For large data centres that means end-to-end management of capacity and data assets, which in turn depends on upgrading the underlying transport.
On the network side, what is needed is greater resource and application awareness, evolving toward SRv6-based network slicing. As operators have trialled and deployed SRv6 in recent years, the industry has put forward the idea of IPv6 plus, giving capacity resources end-to-end on-demand scheduling between data centres. SRv6 is a programmable protocol system built on IPv6 and source routing; its extensibility and programmability make it a strong basis for intelligent slicing, deterministic networking and service chaining.
A computing network is the next stage of cloud-network convergence: it uses distributed coordination across the network to allocate capacity well. As applications move to the cloud, the big internet players have pushed into cloud computing and into edge cloud. China’s three operators have made convergence a core strategy, building on their cloud platforms and their edge coverage to develop MEC (multi-access edge computing) platforms, edge services and private network capability, and extending distributed coordination on top of a programmable network and lightweight cloud-native computing, so that services inside the network can be scheduled sensibly and flexibly.
With the east-west project fully under way, a new pattern is taking shape in which the east and west play complementary roles. As the computing network moves from concept into deployment it needs a solid R&D base, a modern industrial base and real demand behind it, and it needs consensus, collaboration and a healthy ecosystem. Society is already organising around the national hub nodes. Policy is behind the sector and the layout is being set, which is what will put new momentum into a digital economy.
Republished for information. The original text has not been independently verified and does not necessarily reflect our views.
Want to talk through what you just read?
We work end to end, from product design to launch. Tell us about your scenario and we can start there.
Shenzhen XZS Technology Co., Ltd. · Tel 0755-82494862 · 13760424861