Honing in on network innovation and evolution during MWC Shanghai 2024, Toni Eid, Founder of Telecom Review Group & CEO of Trace Media International engaged in an exclusive discourse with Ma Jichun, Chief Expert on Cutting-Edge and Emerging Technologies, China Unicom and Deputy Chief Engineer at China Information Technology Design & Consulting Institute to learn about his experiences and insights.
In the past year, we have witnessed the advent of the intelligent era, and booming technologies such as generative AI (GenAI) and large language models (LLMs) have taken the AI field to the next level. AI is penetrating various industries at an unexpected speed, advocating for a new revolution in productivity and creativity, while redefining every aspect of human society.
Because of this, computing power requirements are increasing sharply, and so is data traffic. This poses higher requirements on and challenges to networks.
Jichun shed light on the fact that building a network that meets AI service development requirements has become a key task for the entire industry. Hence, it is also necessary for telecom carriers, such as China Unicom, to complete the transformation towards technology service providers; it is the inevitable trend of modern productivity.
AI Era: Two Innovative Directions for Networks
Jichun agrees that AI has entered the era of foundation models, and as a result, the demand for intelligent computing power has increased exponentially. "Naturally, the supply of such power has become critical. We need to meet the requirements of foundation model training, as well as for online inference applications everywhere," he mentioned.
The scaled increase of LLM parameters consequently demands higher network bandwidth and stability requirements. Bearing this in mind, as AI computing centers continue to grow, ultra-large-scale or even distributed training tasks will pose even higher requirements encompassing network capacity and performance. AI applications, such as autonomous driving, will also require lower latency and higher reliability.
"In the AI era, I believe network innovation and evolution will go in two directions: one is the evolution of IoT technologies, and the other is network intelligence," pointed out Jichun.
The evolution of IoT technologies includes large-scale networking, ultra-high bandwidth access, wide-area lossless transmission, and large-throughput transmission. These are used for large-scale and distributed training across AICCs, and the wide-area transmission of vast datasets for training purposes.
Moreover, the AI era not only introduces heightened demands for network connection requirements but also necessitates network intelligence. This includes sophisticated network management and control systems capable of supporting intelligent network deployment and configuration.
"It [AI] also means intelligent network devices (devices with native intelligence) can better achieve network programmability, and provide diversified service functions," added Jichun.
Impact of Intelligent Computing
Over the past few years, Chinese carriers have been accelerating the construction of intelligent computing power. Their intelligent computing power is mainly used for intelligent service development and computing power leasing.
In fact, China Unicom's comprehensive computing capabilities form the basis for advancing foundational models, intelligent networks, service transformation, and applications for smart living.
"In terms of computing infrastructure capability, planning and layout, carriers generally use large-scale centralized training centers, combined with edge cloud nodes for integrated training and inference," explained Jichun.
Carriers leverage their extensive backbone networks and MANs to build a next-gen computing network oriented to intelligent computing.
Currently, the majority of intelligent computing power is dedicated to model training. In the short term, the impact of the training traffic is minimal, but Jichun emphasized that in the long term, "new models, like east-west collaboration and cloud-edge collaboration will change the structure of traffic between intelligent computing centers at different levels, while raising new requirements on network traffic scheduling."
On this basis, other related technologies and functions are also evolving. Jichun cited three examples to illustrate:
- Wide-Area Congestion Control: Supports lossless data transmission and assurance, making computing power connections more efficient.
- High-Throughput Transmission: Entails multi-channel transmission through intelligent scheduling over wide-area networks, facilitating rapid data exchange and staggered data transmission.
- Computing Power Scheduling: Leverages an automatic scheduling system, utilizing algorithmic innovations to swiftly connect computing resources, rapidly provision services, and intelligently adjust user access.
Building Computing Networks for B2B Competitiveness
B2B plays an increasingly important role in carriers' operations, especially now that enterprises are accelerating their digital transformation journey by embracing AI.
"To make their B2B services more competitive, carriers need plans for leading foundation,
technological innovation, and differentiated products," noted Jichun. "In terms of foundation and technological innovation, China Unicom is an active promoter of integrating computing power centers and the Internet with intelligent scheduling."
China Unicom has successfully constructed a multi-level computing power center in the country and has built an industrial Internet and intelligent metro network. Notably, these computing networks function utilizing new technologies and architectures.
As for intelligent scheduling, China Unicom employs advanced computing-network integrated scheduling. To improve cloud network and computing network product competitiveness, the telco offers a variety of cloud-network synergy products.
For example, China Unicom offers cloud interconnection products with multi-cloud connections and online provisioning. The telco's secure private line products are based on leading cloud network security capabilities, while its cloud-network convergence products are penetrating campus networks and IoT fields. These offerings aid China Unicom in achieving greater market-share.
"As AI develops, China Unicom continues to bring it into our cloud-network synergy products as we continue to consolidate our competitiveness," concluded Jichun.
Conclusion
AI brings significant changes to networks in terms of performance, quality, and user experience, necessitating breakthroughs in networking technologies and network intelligence.
As a leader in the information communications service and intelligent computing industry, China Unicom has been preparing for this transformation by building high-quality cloud networks and computing networks. This has allowed the telco to achieve differentiated competitiveness and advantages across the B2B segment.
Jichun's overall analysis and China Unicom's successful use cases are a testimony to the efforts the telco is taking to enable ubiquitous computing power and convert the technologies associated with AI from a pipedream into a reality.
