The development of the AI industry is in full swing, the fierce battle between various types and thousands of models is in full swing, the scale of computing power clusters is growing exponentially, and the connection methods between AI and industries and users are also diverse. In such a competitive background, the integrated development of computing power and network is becoming the focus of attention of the industry and users. Zeng Fugui, Senior Vice President and President of Network Product Line of H3C Group, said: "At a time when the demand for computing power is unprecedentedly prosperous, the diversification of the computing power ecosystem has become an inevitable trend, and the innovation of connection technology should also complement it. On the one hand, the network should have the ability to open standards to solve the problem of computing power expansion, so that thousands of computing cards can be turned into a whole, and then the quantitative change of computing power can promote the qualitative change of large models. ”
Zeng Fugui, Senior Vice President of New H3C Group and President of Network Product Line
With the application and AI two-wheel drive, AD-NET has entered the 7.0 era
Application-Driven Network (APIN) is a technology concept that H3C has practiced in the network field for many years, and based on this, it has created many key products covering three major scenarios: data center, campus network, and WAN. In the past few years, it has embodied the core idea of H3C attaching importance to business applications and user experience in the digital era, and in the new intelligent era, H3C is upgrading the connotation of AD-NET to Application Driven + AI Driven through the comprehensive application of AI and various advanced technologies, with "double A driven" Realize the multiplication of user revenue and build an intelligent foundation for the industry in the new era.
In order to meet the widespread demand for intelligence, Xinhua III has established a set of positive cycles of "network, computing, and intelligence" with new technologies, new products, and new ideas. AD-NET 7.0 is fully equipped with AI capabilities through the built-in Lingxi large model, combined with the NAI intelligent native technology embedded in H3C network equipment, to help users enjoy the multiple conveniences brought by AI technology in advance. As a result, the dual-A drives of Application Driven and AI Driven can also power each other in users' business scenarios, bringing users longer-term and stronger value benefits.
Data center: Build an open intelligent computing network for heterogeneous GPU intelligent computing clusters
The network is the key to building an AI computing center/cluster, but traditional data centers face great technical challenges.
On the one hand, one-to-many and many-to-many data flows often occur in the system, and the network packet loss caused by them is difficult to avoid by simply increasing the port bandwidth, which greatly hinders the further increase of the cluster scale. Many users have no choice but to use the more closed and expensive Infiniband network to avoid packet loss and cluster performance loss.
On the other hand, more and more manufacturers have begun to launch GPU accelerator card products, and different GPUs will use different communication libraries, and Infiniband cannot be fully compatible with GPU products from different manufacturers due to its closed nature.
Faced with the dilemma, H3C has three solutions: first, "expanding the circuit and increasing capacity", that is, further pushing up the port bandwidth of the network, so that the network has the ability to cope with larger-scale data transmission; second, "splitting incremental", that is, breaking the chassis limitations of switching equipment, allowing users to obtain more ports through various networking forms, and then forming a larger computing power cluster; and third, "balancing and increasing efficiency", that is, innovatively introducing DDC (Disaggregated) with cell-level load balancing capabilities Distributed Chassis network architecture, which decouples software and hardware while achieving efficient load balancing, makes the performance of the intelligent computing center to a higher level. At the same time, H3C launched a new intelligent computing network solution.
First of all, the intelligent computing network solution provides users with switch products with different port densities and rich forms, which can cope with larger-scale burst traffic and increase the bandwidth limit. Second, the new solution can provide a variety of networking forms, such as single-frame single-layer, box-box two-layer, and box-box two-layer, which can provide more ports for intelligent computing centers and clusters, which is more conducive to cluster expansion. Third, the DDC network architecture adopted by the core product H3C S12500 AI series switches can provide cell-level load balancing capabilities to ensure 100% lossless transmission, combined with different load balancing technologies such as LBN port symmetric load balancing, DLB dynamic stream-by-flow load balancing, FGLB global dynamic load balancing, and SprayLink dynamic packet-by-packet spraying, which can fully meet the network optimization and traffic balancing goals of different intelligent computing scenarios. Regardless of whether the node supports SmartNIC or not, the new solution can completely solve the packet loss problem caused by network congestion.
As a result, the new solution can be based on a standardized Ethernet technology stack, maintaining an open and compatible architecture, while also solving the problem of cluster efficiency, paving the way for further growth of computing power.
Campus network: Build an all-optical network that is intelligent, fast, and simplified based on access experience
H3C is a participant and promoter of the all-optical trend of campus networks. Through the Ethernet all-optical + PON convergence technology path, it not only greatly improves the user bandwidth at the access layer, but also further reduces the energy consumption and TCO of the campus network, and improves the service life of the entire network. On the other hand, H3C saw that the technological development process of the campus network was actually ahead of the rigid needs of applications for network performance, so it introduced more innovative AI technologies into the campus network, improved the efficiency and experience of O&M management through more refined granularity, and built a smart and simple campus network.
Based on AI technology, H3C can visualize the end-to-end application experience of the campus network. As a result, O&M personnel can use the dynamically updated digital map of the campus network to actively perceive the access and network usage of a large number of terminals and services in the network, ensuring continuous improvement of the final application experience.
In terms of simplified management, H3C has launched a lightweight campus BRAS (Broadband Remote Access Server) and a Central AC solution integrated with wireless 4i technology, which can greatly simplify the management complexity of wired and wireless user policies, reduce the workload of O&M, and achieve on-demand campus policy management and consistent campus user experience.
In the field of access, on the one hand, H3C is actively promoting the launch of new products of FTTD access products, scenario-based Wi-Fi 7 APs and industrial switches, providing users with more diverse access layer devices and improving the scenario adaptability of all-optical networks.
Through multi-dimensional innovations in O&M, management, access, and cabling, H3C's campus network solution can further improve the operation, management, and network use experience of all-optical campus networks, promote the deep integration of computing power and services, and make intelligent experiences reach edge services and everyone along the network cable and Wi-Fi.
WAN: 400G high-speed WAN is used to realize intelligent sharing of computing power
The demand for computing power can appear in hundreds of industries, but the construction of intelligent computing centers has many limitations. Therefore, how to connect the intelligent computing center and business needs in the WAN mode has become a key area of concern for users. H3C believes that bandwidth, algorithms, and reliability are still the key points in building a wide-area computing power network.
Taking the main core routing products such as CR19000 and CR16000E-F as examples, H3C has made three upgrades:
The first is to provide higher 400G forwarding rates and dramatically reduce WAN latency and jitter with deterministic network technology. By using technologies such as DetNet and DetNetOAM, H3C routers can achieve ultra-low transmission latency of 1 millisecond in metro, 5 milliseconds in region, and 20 milliseconds in core, as well as network jitter as low as 15 microseconds, greatly improving the quality of computing power networks.
Second, H3C also integrates the computing power factor into the routing algorithm embedded in the network equipment, so that the WAN is naturally suitable for transmission computing power.
Third, it allows users to build dedicated computing power channels on demand and provide users with service-oriented computing power dedicated lines. Through a series of features such as parameter selection, on-demand construction, automatic network construction, dismantling when used up, and dynamic bandwidth adjustment, H3C routers can further improve the resource utilization and network SLA of the computing power network.
Under the triple evolution, cross-domain computing power scheduling can enter the real practical stage, allowing users to obtain intelligent computing power at a lower cost, more freely and flexibly, and deploy intelligent computing services, so as to achieve inclusive computing services and enable the computing industry to gain a broader development space.
Intelligent O&M upgrade: Use large models to address multiple challenges
At a more advanced level of O&M management and security, H3C is actively promoting the deep integration of large-scale model technology with traditional ICT operation and management and security solutions. The Lingxi model is the crystallization of H3C's decades of O&M experience in the ICT field and the wisdom of tens of thousands of experts.
Taking network O&M management as an example, the Lingxi assistant supported by the Lingxi model can integrate the operation status data of the entire network to achieve a series of functions such as AI visualization, AI troubleshooting, AI tuning, and AI security, lowering the threshold for network O&M and security work, reducing the manual operation workload of O&M personnel, improving the level of network automation, and liberating users from complicated basic work.
In addition to intelligent O&M, Lingxi Assistant also integrates functions such as knowledge Q&A, configuration guidance, and product recommendation, which can help users better understand and use H3C products and provide reference for the continuous development and evolution of network architecture.
Build an intelligent foundation for the industry with high-quality network connectivity
Whether it is the information age in the past, or the digital and intelligent era today, whether it is the construction of cloud computing centers, or today's intelligent computing clusters and intelligent computing centers, whether in data center networks, campus networks, and WANs, the network is connected to computing power and storage power, services and users. As the infrastructure becomes larger, more powerful, and more complex, network technologies must evolve simultaneously or even exceededly, so that the entire ICT architecture can meet expectations in terms of management, efficiency, and reliability, and provide users with a service experience that matches the times.
As a promoter of network evolution, H3C needs to continue to expand the upper limit of network connectivity with new technologies, new ideas, and new solutions, and consolidate loose ICT elements into an intelligent foundation to lay a solid foundation for the upward development of user services. This road is full of challenges and never ends, but Xinhua III has always strived to move forward and strive for the top.