Search

Huawei, IDC, and SenseTime Jointly Release Data Center Network Evolution White Paper for the AI Era

2024.09.20

[Shanghai, China, September 20, 2024] During HUAWEI CONNECT 2024, Huawei collaborated with IDC and SenseTime to unveil the White Paper on Data Center Network Evolution in the AI Era (hereinafter referred to as the White Paper) at the data center network session themed "Xinghe Intelligent Fabric, Powering the AI Era". This document outlines the evolving trends and challenges of data center networks in the AI era, highlighting the technical leadership and impact of Huawei's Xinghe Intelligent Fabric solution across three key areas: AI service development trends, intelligent computing solutions, and industry use cases.

Huawei, IDC, and SenseTime releasing the White Paper on Data Center Network Evolution in the AI Era

Huawei, IDC, and SenseTime releasing the White Paper on Data Center Network Evolution in the AI Era

The White Paper indicates that the rise of generative AI has created higher demands for data center networks, with requirements that far exceed those of traditional networks' changes and linear traffic growth. As AI models scale, the growth in model parameters and training datasets necessitates larger cluster sizes and enhanced computing capabilities. To fully harness the potential of these computing clusters, the network must effectively coordinate the entire system to optimize performance across all functional areas of the AI model. Furthermore, as network scale expands, the number of fault domains increases, raising operations and maintenance (O&M) requirements.

In response to these challenges, Huawei has introduced the Xinghe Intelligent Fabric solution. The network scale load balancing (NSLB) algorithm boosts network throughput from 50% to 95%, enhancing AI training efficiency by over 10%. Huawei's optical module fault detection and channel loss resistance technologies reduce the annual interruption rate of AI training by 90%. Additionally, Huawei's liquid-cooled fixed-form Ethernet data center switches achieve a 1:5 air-to-liquid ratio, resulting in approximately 85% savings in air conditioning power consumption, with an optimal Power Usage Effectiveness (PUE) of less than 1.15. The Huawei NetMaster network large model along with the exclusive network digital map enables AI-driven network detection, health evaluation, and communication exception diagnosis, significantly improving O&M efficiency.

Looking ahead, Huawei remains committed to collaborating with partners and customers to advance technical research and innovation in the data center network domain. Also, Huawei aims to continuously optimize network architecture and deliver more efficient and intelligent data center network solutions across various industries, facilitating their digital and intelligent transformations.

TOP