High-Throughput Ethernet ETH+ Protocol Released: Building an Open Intelligent Computing Network for the AI Large Model Era

High-Throughput Ethernet ETH+ Protocol Released: Building an Open Intelligent Computing Network for the AI Large Model EraThe explosive growth of AI technology has led to a surge in demand for computing power. This data-driven intelligence revolution has not only injected new vitality into the development of high-performance computing technologies but also posed higher requirements for algorithm optimization, hardware upgrades, and system architecture

High-Throughput Ethernet ETH+ Protocol Released: Building an Open Intelligent Computing Network for the AI Large Model Era

The explosive growth of AI technology has led to a surge in demand for computing power. This data-driven intelligence revolution has not only injected new vitality into the development of high-performance computing technologies but also posed higher requirements for algorithm optimization, hardware upgrades, and system architecture. However, traditional Ethernet data center network architectures are built based on general computing scenarios, making it difficult to meet the needs of AI for high-speed, stable, and scalable computing networks. To address this challenge, the High-Throughput Ethernet Alliance, consisting of over 40 organizations including Alibaba Cloud and the Institute of Computing Technology of the Chinese Academy of Sciences, released the 1.0 version of the High-Throughput Ethernet ETH+ protocol at the CCF National High-Performance Computing Academic Conference on the morning of September 24th.

 High-Throughput Ethernet ETH+ Protocol Released: Building an Open Intelligent Computing Network for the AI Large Model Era

Cai Dezhong, Co-Chairman of the High-Throughput Ethernet Alliance, Vice President of Alibaba Cloud Research and Development, and Head of Infrastructure Network, pointed out in an interview that traditional Ethernet data center network architectures can only provide "best effort" network performance, failing to meet the requirements for high-performance, high-stability, and high-scalability computing networks in the AI era. The launch of the ETH+ protocol aims to solve this problem, providing a more solid network foundation for the AI large model era

 High-Throughput Ethernet ETH+ Protocol Released: Building an Open Intelligent Computing Network for the AI Large Model Era

The ETH+ protocol optimizes the frame format, achieving a 74% increase in payload ratio, significantly improving data transmission efficiency. Meanwhile, through deep support for link layer and physical layer retransmission technologies, the semantic reliability of ETH+ Ethernet has been significantly enhanced. Additionally, ETH+ leverages RDMA in-network computing technology to achieve a 30% improvement in collective communication performance, effectively addressing the shortcomings of traditional Ethernet and enabling it to better meet the needs of the AI era for efficient, stable, and scalable computing networks.

 High-Throughput Ethernet ETH+ Protocol Released: Building an Open Intelligent Computing Network for the AI Large Model Era

The High-Throughput Ethernet Alliance was jointly initiated by Alibaba Cloud and the Institute of Computing Technology of the Chinese Academy of Sciences. Members include Peking University, Pingtouge, Shengke, Tencent, ByteDance, ZTE, H3C, Yunhe Zhiwang, Biren, and over 40 other organizations. At the release ceremony, Sun Ninghui, a member of the Chinese Academy of Engineering and chairman of the alliance, and Cai Dezhong, Vice President of Alibaba Cloud Research and Development, together with Chen Guoliang and Wu Jiangxing, both members of the Chinese Academy of Sciences, and representatives of core members of the alliance, jointly unveiled and released the 1.0 version of the High-Throughput Ethernet ETH+ protocol standard.

The underlying basis of intelligent computing networks is the network protocol. Unlike private commercial protocols tightly bound to hardware, the Ethernet protocol has gained widespread industry support due to its open and collaborative nature. Cai Dezhong stated that the alliance brings together forces from industry, academia, research, and application, aiming to develop and establish unified technical standards and foundations for intelligent computing networks based on the open Ethernet ecosystem. The establishment of the alliance has not only promoted cooperation and exchanges between domestic chip companies but also driven technological innovation and achievement transformation.

Several chip companies have already implemented product development and deployment based on the ETH+ protocol. It is expected that a batch of clusters based on domestic network chips will emerge early next year. A series of achievements, including open-source hardware IPs, commercial IPs, verification systems, and benchmark test sets matching the ETH+ protocol, have been presented. Among them, the basic network layer IP based on the ETH+ protocol has achieved widespread commercialization, and the open-source RDMA NIC design has provided industry reference.

Cai Dezhong is confident about the future development of the ETH+ protocol. He believes that with the participation and joint efforts of more industry, academia, research, and application units, the ETH+ protocol will have a profound impact globally, contributing Chinese strength to the development and prosperity of intelligent computing network technologies and ecosystems.

The release of the High-Throughput Ethernet ETH+ protocol marks a significant step forward for Ethernet technology in the field of intelligent computing networks. In the future, as the ETH+ protocol continues to develop and improve, it will build a stronger, more open, and more efficient intelligent computing network foundation for the AI large model era, driving greater breakthroughs in AI technology.


Disclaimer: The content of this article is sourced from the internet. The copyright of the text, images, and other materials belongs to the original author. The platform reprints the materials for the purpose of conveying more information. The content of the article is for reference and learning only, and should not be used for commercial purposes. If it infringes on your legitimate rights and interests, please contact us promptly and we will handle it as soon as possible! We respect copyright and are committed to protecting it. Thank you for sharing.(Email:[email protected])