The network pulse of new H3C: pumping blood for AI and rushing towards industry

As the latest general-purpose technology, AI large models have been developing in full swing since this year. Many practitioners and experts have also noticed that during the training and application process of AI models, network upgrades and adaptations need to be given priority.

If data centers and computing power clusters are the "heart" of AI, then the network is like the "artery" of AI, responsible for the "blood circulation" of production resources such as computing power and data.

For the application and implementation of AI technology, AI computing power must be available at any time like water and electricity. To achieve this goal, network connection is a core element.

In recent years, various infrastructure service providers have also been making intensive arrangements for the widespread application of AI. Among them, H3C Group has its own systematic thinking and laid out many technologies in response to the AI ​​​​large model and AIGC boom, relying on more than 20 years of profound accumulation in the field of enterprise networks.

The network foundation, technical system, and service model that are compatible with AI computing power. We start from the thinking of New H3C to understand the pulse of China in the intelligent era.

Cultivation of internal strength: blood pumping mechanism of AI large model

The large AI model represents the latest digital productivity. We start from the AI ​​mechanism to understand why the connection force represented by the network is an indispensable "internal force" in the AI ​​era.

Let me ask a question first: Is the current bottleneck of large AI models the inability to buy GPU cards?

No, it is network transmission.

The first is hematopoiesis - training large models, the network is the aorta of the "AI heart".

An internal report from OpenAI mentioned that cross-server data transmission is the biggest bottleneck in training very large models.

This is because large AI models usually need to be deployed on AI computing clusters to achieve training and inference acceleration. In order to reduce training time, multi-GPU distributed training is usually used. Parallel training on multiple devices must solve the communication bottleneck between servers, which also puts forward higher requirements for network bandwidth performance.

Traditional networks are not designed for the large traffic, ultra-high throughput, and ultra-low latency transmission required by large AI models. They can only provide "best effort" connections. In the face of this wave of AIGC boom, It may cause problems such as massive delays, unstable transmission rates, and excessive operation and maintenance costs.

If the AI ​​computing power cluster is the "heart" of a large model, and the network, as the "aorta", must be expanded to pump out the ultimate computing power.

The second is blood transfusion - using a large model, the network is the "blood vessel" reaching the end.

AI large models and AIGC applications will eventually reach hundreds of industries and be transformed into actual productivity. From the "heart" of the computing cluster to the "limbs and limbs" of application scenarios in various industries, network requirements are more complex and stringent.

In the context of computing in the East and in the West, the concentration of computing power allows enterprises to obtain computing power at a lower cost. At the same time, a cross-regional wide-area network is needed to complete large-capacity and sudden business data transmission. , and the coordination and scheduling of computing power. At the same time, as enterprise business migration to the cloud becomes mainstream, production equipment is increasingly connected to the Internet through IP. AI applications ultimately rely on a large number of IoT devices to be effective, such as smart cameras and sensors. Therefore, the transmission of AI also requires IP networks. taken into consideration.

AI scenarios are applied in every corner, such as streets, highway checkpoints, etc. AI model applications and industrial digitization must directly reach the application layer. IP network is the most widely used network solution and is difficult to replace by other networks.

Only when the network system is fully upgraded can enterprises make good use of AI large models and be reborn through digitalization and intelligence.

Then comes management - optimal network energy efficiency and healthy development of industrial AI.

Just as the human body cannot always be excited, it will automatically adjust its body functions according to the intensity of exercise to maintain long-term health, and the same is true for AI. The new AI infrastructure is a large-scale project that spans the east and west and supports the digitalization of various industries. It will bring huge investment in network iterations, system maintenance, etc. How to balance energy consumption, operation and maintenance and other costs is also a question that needs to be answered.

It can be seen that the network system and architecture in the AI ​​era are more complex and diverse. To accurately and losslessly deliver the technical dividends and computing power value of AI to all walks of life has brought great challenges to the network.

Faced with the imperative supply-side reform of the network, New H3C has found a path to “cut the sinews and cleanse the marrow.”

Opening up the meridians: New H3C’s network supply-side reform

Readers who are familiar with martial arts novels know that even if one has strong internal strength, narrow meridians can easily lead to loss of control. In the AI ​​era, in order to make full use of computing power and applications, we must allow them to flow quickly on an unprecedentedly powerful network. A supply-side reform of network connectivity is imperative.

Zeng Fugui, vice president of H3C Group and president of the network product line, shared five technology trends in the network field in the era of artificial intelligence. We can see that the network supply-side reform is a relevant process that requires continuous training.

(Vice President of New H3C Group and President of Network Product Line Zeng Fugui)

The first step is to quickly iterate the business.

The digital transformation of all walks of life is accelerating, and some new industry tracks have been expanded, such as the industrial Internet, connected cars, etc. The digitalization and intelligent upgrading of traditional industries are the core needs of the network supply-side reform and also the technology The starting point for upgrade iterations.

Connectivity, as a productive force, enables digitalization and intelligent industries in various industries, which requires a higher-quality network.

Includes high reliability to ensure the stable operation of a large number of digital services; high performance to support the transmission and processing of massive data; high security to protect enterprise business and data from attacks and leaks; flexibility to support rapid deployment and adjustments to meet the needs and changes of different businesses; scalability to meet the expansion needs brought about by rapid business growth; and the ability to automate to reduce operation and maintenance costs and improve operation and maintenance efficiency.

H3C continues to upgrade network technology solutions to meet the connectivity needs of various industries. For example, the industry's first IP-optical convergence technology greatly simplifies the network architecture through the integration of IP equipment and transmission equipment, and uses the unification of multi-layer network path calculation, control, and maintenance to avoid increasingly complex networks restricting business development. The all-optical network solution provides a new generation of campus network solution with Ethernet all-optical and PON all-optical integration for campus, medical, enterprise and other campus scenarios, making the implementation of digital services in the campus more worry-free and labor-saving.

In general, network upgrade and adaptation are the core elements of a new round of industrial digitalization and intelligence.

The second step is to cloudify all services.

When it comes to industrial digitalization and intelligence, migrating business to the cloud is a shortcut. With the in-depth cloudification of various industries, all businesses of enterprises will be migrated to the cloud to achieve cloud-based management and operations. The cloud transmission and processing of large amounts of data, the stable production of remote collaboration, and the terminal implementation of a large number of intelligent applications have put forward extremely large bandwidth connection requirements for the wireless network at the cloud wide area network access end.

On the transmission side, a high-quality wide area network is required. New H3C uses SRv6 programmability and FlexE hard channel conversion capabilities to use deterministic network technology to ensure the certainty of cross-domain transmission in wide area networks. Currently, it can control the delay jitter within 1,600 kilometers of wide-area transmission distance. Within microseconds, the problem of coexistence of deterministic traffic and non-deterministic traffic on the same link is overcome, and the service level can reach thousands. For example, industrial Internet, telemedicine, smart driving and other network connection scenarios that require high quality and low latency can be well satisfied.

On the terminal side, an access network with wide coverage and large bandwidth is required. After business cloudification, it will drive the digitization of a large number of terminal equipment and services. IoT devices will be intensively connected, and the demand for access-side bandwidth will increase. New H3C continues to explore wireless connection technology and is at the forefront of the industry, launching Wi-Fi 7 products with larger bandwidth and lower latency, and providing more reliability, high density and low latency at the underlying technology and protocol levels. Latency guarantee, compared to Wi-Fi 6, can make wireless and wired access experiences truly consistent. For example, unmanned vehicles in smart terminals, AGV cars in black light factories, automatic patrol robots in parks, etc. These smart applications can all be based on Wi-Fi 7 ultra-wide networks to provide services more quickly and reliably.

New H3C deepens the application of new technologies to meet the needs of full-service cloudification, accelerates the commercial transformation and implementation of new technologies, improves productivity, and activates the potential of the workforce.

The third step is to fully apply IPv6.

Moving business to the cloud requires more IP addresses and better network connections to ensure that services can be delivered in real time. IPv4 addresses have long been exhausted, and IPv6 networks have become the general trend. The "Notice on Accelerating the Large-Scale Deployment and Application of Internet Protocol Version 6 (IPv6)" clearly emphasizes the key direction of promoting the large-scale deployment and application of IPv6 by 2025. Among them, the focus of application expansion is to deepen the integration of IPv6 technology and industry.

In compliance with policy guidance, H3C will continue to promote the upgrade of IP networks through deterministic network and service chain technologies, provide more reliable, more sophisticated and more flexible guarantees for IPv6 applications, evolve to IPv6, and continue to empower digital services.

For example, the all-scenario router SR6608-M can support high-quality video calls, streaming entertainment and other applications for C-end users, providing a more stable and reliable network connection for mobile devices. With the deterministic network created by IPv6, B-side scenarios such as smart cities and smart agriculture can provide more IP addresses, realize large-scale equipment and sensor connections, establish a safe, manageable and visible network environment, and ensure business certainty. sexual experience.

The fourth step is the rapid development of AI and the explosive growth of practical applications.

With the digital base, cloud services are becoming more and more abundant, and AI applications are expected to be quickly integrated into industrial scenarios, serving as an engine for industrial upgrading, improving productivity and reducing labor costs.

Cheng Zhen, general manager of the Network Product Line System Planning and Solutions Department of New H3C Group, mentioned: “Large models like GPT 4 reach one trillion parameters. In the process of obtaining data, traditional computing clusters suffer from packet loss, Due to problems such as delay, there is no way to fully utilize the computing power. The solution is to increase the communication bandwidth between nodes to ensure the overall computing efficiency."

To this end, H3C has laid out a large number of basic technologies in advance, such as CPO silicon light engine, multi-level non-blocking technology, cold plate liquid cooling technology, etc., which can meet the needs of ultra-large-scale computing power and massive data transmission in data centers. High-speed transmission can also take into account the requirements of green energy saving.

It is particularly worth mentioning that the next-generation ultra-wide data center network based on CPO silicon light engine technology, combined with effective data compression and calculation algorithms, can improve the transmission efficiency of the computing system and achieve efficient and lossless data transmission to ensure data transmission without packet loss and low latency, meeting the extreme computing needs of intelligent computing scenarios.

This series of technologies helps data center networks continue to innovate and upgrade. As computing power grows explosively, industry users can quickly seize large model opportunities and join the new round of AI boom.

The fifth step is to deepen the integration of network and application.

Under the above technology trends, the services carried by the network are becoming more and more abundant, and the difficulty of management is naturally increasing. For example, a large number of enterprises are moving more data, business, and equipment systems to the cloud to achieve 24/7 online business, which puts forward new requirements for the trustworthy security capabilities of the network. The network architecture based on the zero trust principle is often very complex, which brings a lot of inconvenience in operation and maintenance management. How to take into account the convenience of operation and maintenance management while achieving zero trust security and credibility in the future has become a necessary test point for network solutions.

In the eyes of H3C, the network is not only an unchanging connection "pipeline", but also is deeply integrated with applications, changing with application trends and changes, and becoming a flexible, intelligent, and autonomously evolving service.

New H3C has injected AI elements into its intelligent connection solution, completed the automation and intelligent upgrade of the entire life cycle, and provided customers in various industries with a smart network with simplified delivery, intelligent operation and maintenance, and the ultimate experience.

For example, when a high-traffic application occurs, the autonomous network will automatically allocate more bandwidth and resources to ensure that application performance is not affected. At the same time, network management will automatically adjust according to changing application requirements, such as stricter security control of highly sensitive data; dynamic monitoring of the operating status of computing equipment, etc., such as high-speed operation of large AI computing models. When the flow is not that large, the cooling equipment such as fans and power supplies are operated at full capacity; when the flow is not that large, real-time dynamic peak shaving is performed, the fans are slowed down, and the power supply is adjusted to achieve a better energy efficiency ratio.

Data is the core asset of an enterprise, and security issues are a top priority. New H3C's new security platform uses AI to empower security detection, intelligently identify and isolate network attacks and virus detection, and achieve rapid response. At the defense level, passiveness has been transformed into proactiveness. Unlike the past where patches were applied after security issues were exposed, AI analysis is used to judge trends, and the system will provide early warning of risks and issue corresponding strategies. At the same time, the attack source is modeled and source traced, and the cloud and network are linked to accurately block the attack source and nip it in the bud.

It can be seen that AI is just an opportunity and a breakthrough. New H3C has opened up the "Rent and Supervision" of the network and realized the integration of AI from computing power, data, application to business and scenarios. Only then can the "digital productivity" be transformed. The "prehistoric power" is released to enable digitalization and intelligence in various industries.

One sword flying fairy: Smart China, running on the main artery of the Internet

What will happen when the powerful internal power of the large AI model is fully released through the new H3C network?

We all know that in the past three industrial revolutions that have changed the world structure, China has been an underachiever and a learner. The fourth industrial revolution is a once-in-a-century change that China cannot miss. The greatest possibility of the fourth industrial revolution is the intelligent revolution represented by AI technology.

There is a consensus in China: if you want to be rich, you must first build roads. As an important part of the new AI infrastructure, the network is the main artery of digitalization and intelligence. Only when this "main artery" is smooth can the full implementation of AI and the continuous advancement of industrial intelligence be achieved.

Today, the upsurge of intelligent industrialization with AI large models as its core has reached the eve of its outbreak. The new H3C intelligent connection network is also ready. When the "internal power" of AI and the "arteries" of the Internet complete the intersection in China, it may be the "flying fairy" we have been waiting for for a long time.

In the complete industrial solution brought by the new H3C, we can see some possibilities:

First is completeness. New H3C has built a full-stack technology system from AI training to application, operation and maintenance, from the backbone network to the access network, supporting the implementation of AI in different scenarios such as cloud, pipe, edge, and terminal. New H3C's network , can be said to be the power source of intelligent connection in all scenarios and industries.

The second is advanced. New H3C has accumulated technological advantages in the network field to achieve a comprehensive upgrade of connectivity. The changes brought about are disruptive. It can support new applications such as AI large models, autonomous driving, industrial Internet, and campus networks, and activate hundreds of new applications. Intelligence surges across all industries.

Finally it’s very simple. New H3C will transform the network from "hard power" to "soft service", making network upgrades low-threshold, low-cost, and high-return. It will also drive the enthusiasm for industrial digitalization and intelligence, integrating into the fourth wave The torrent of the industrial revolution.

These changes are happening bit by bit on the land of China, which reminds me of a poem called "Life" by the poet Bei Dao. The whole text has only one word: net.

Today, with the rapid development of information technology, our world is running on tangible or invisible networks. Along with the network laid by the new H3C, all walks of life are leading to the intelligent era, and intelligent life is coming.

As Zeng Fugui said: This kind of deep AI represented by ChatGPT will lead the next wave of technology and revolution, bringing about earth-shaking changes to all walks of life, including the work and life of each of us. .

The next stop of H3C, the future of intelligent connection and the picture of the fourth industrial revolution, is among the lights of smart China.

The above is the detailed content of The network pulse of new H3C: pumping blood for AI and rushing towards industry. For more information, please follow other related articles on the PHP Chinese website!