面向跨智算集群互联的新型HIC-OTN技术

张德朝; 孙将; 曹珊; 左铭青; 王东; 李晗

doi:10.11959/j.issn.1000-0801.2025117

您当前的位置：

首页 >

文章列表页 >

面向跨智算集群互联的新型HIC-OTN技术

专题：智算光互联 | 更新时间：2025-10-31

- 面向跨智算集群互联的新型HIC-OTN技术
- Novel HIC-OTN for interconnection of cross-intelligent computing clusters
- 电信科学 2025年41卷第4期页码：53-60
- 作者机构：
  
  中国移动通信有限公司研究院，北京 100053
- 作者简介：
  
  [ "张德朝（1979- ），男，博士，中国移动通信有限公司研究院基础网络技术研究所副所长、正高级工程师，主要从事 SPN/PTN、OTN/WDM、前传open-WDM、PON 和同步等光传输与接入网领域的技术研究和标准化工作。" ]
  [ "孙将（1987- ），男，博士，中国移动通信有限公司研究院工程师，主要从事OTN/WDM、前传open-WDM的研究工作。" ]
  [ "曹珊（1990- ），女，博士，中国移动通信有限公司研究院项目经理，主要从事高速大容量光通信技术及B1T电层技术的研究工作。" ]
  [ "左铭青（1995- ），男，博士，中国移动通信有限公司研究院工程师，主要从事高速大容量光通信技术研究工作。" ]
  [ "王东（1986- ），男，博士，中国移动通信有限公司研究院技术经理、主任研究员、高级工程师，主要从事400 Gbit/s/超400 Gbit/s OTN/WDM、智算无损OTN、前传open-WDM领域研究和标准化工作。" ]
  [ "李晗（1975- ），男，博士，中国移动通信有限公司研究院基础网络技术研究所所长、教授级高级工程师，主要从事SPN/PTN、前传open-WDM、OTN/WDM、PON和同步等光通信领域研究工作。" ]
- 基金信息：
  
  国家重点研发计划项目(2024YFB2908305)
- DOI：10.11959/j.issn.1000-0801.2025117
  中图分类号： TP393
- 收稿：2025-03-04，
  
  修回：2025-04-15，
  
  纸质出版：2025-04-20
- 稿件说明：
移动端阅览
张德朝,孙将,曹珊等.面向跨智算集群互联的新型HIC-OTN技术[J].电信科学,2025,41(04):53-60.

ZHANG Dechao,SUN Jiang,CAO Shan,et al.Novel HIC-OTN for interconnection of cross-intelligent computing clusters[J].Telecommunications Science,2025,41(04):53-60.
张德朝,孙将,曹珊等.面向跨智算集群互联的新型HIC-OTN技术[J].电信科学,2025,41(04):53-60. DOI： 10.11959/j.issn.1000-0801.2025117.

ZHANG Dechao,SUN Jiang,CAO Shan,et al.Novel HIC-OTN for interconnection of cross-intelligent computing clusters[J].Telecommunications Science,2025,41(04):53-60. DOI： 10.11959/j.issn.1000-0801.2025117.

摘要

随着全球 AI 产业蓬勃发展，大模型技术对算力需求越来越高，国内外大型科技公司正积极投建超万卡/超十万卡集群。超十万卡集群的发展受限于水电供应、建设投入等因素，而通过高速全光网构建多集群互联的基础网络底座，实现跨集群高效协同训练是重要的潜在解决方案。为满足跨智算集群互联超大带宽、超低时延、超高可靠需求，提出了无损智算光传送网（hitless intelligent computing optical transport network，HIC-OTN）的技术架构及关键技术方案，并基于HIC-OTN完成了首个104 km跨智算集群流水线并行（pipeline parallelism，PP）拉远训练现网技术试验，探索和验证了百公里级跨集群PP训练的可行性。基于800 Gbit/s HIC-OTN互联在52 km和104 km两个智算集群场景下，HIC-OTN完成了等同单节点训练效率98%以上的高效协同训练，并实现了光网络保护倒换对训练效率的无损和无感知。

Abstract

With the rapid development of the global AI industry

the computational power demands of large-scale models continued to grow

prompting major technology companies worldwide to actively construct ultra-large-scale clusters exceeding 10 000 or even 100 000 GPU. Limited by natural resource supply

construction investment

and other constraints

the construction of a multi-cluster interconnected fundamental network through a high-speed all-optical network is an important potential solution for achieving efficient collaborative training across clusters. To meet the ultra-large bandwidth

ultra-low latency

and ultra-high reliability requirements of intelligent computing interconnection

a hitless intelligent computing optical transport network (HIC-OTN) and its key technological solutions were proposed. Based on HIC-OTN

the first field trial of 104 km cross-cluster pipeline parallelism (PP) training had been demonstrated

verifying the feasibility of 100 km-class cross-cluster PP training. Based on the 800 Gbit/s HIC-OTN interconnection

highly efficient collaborative training was achieved in two scenarios (52 km and 104 km clusters)

delivering over 98% of the single-node training efficiency. Moreover

hitless and imperceptible optical network protection switching was demonstrated

ensuring zero impact on training performance.

关键词

Keywords

references

DISKIN M , BUKHTIYAROV A , RYABININ M , et al . Distributed deep learning in open collaborations [C ] // Proceedings of the 35th International Conference on Neural Information Processing Systems . New York : Curran Associates Inc. , 2021 : 7879 - 7897 .

中国软件评测中心 . 人工智能大语言模型技术发展研究报告（2024年） [R ] . 2024 .

China Software Evaluation Center . Research report on the development of artificial intelligence large language model technology (2024) [R ] . 2024 .

丁宏庆 , 张鹏飞 , 牛红韦华 , 等 . 云化的智算中心万卡集群创新与实践 [J ] . 电信科学 , 2024 , 40 ( 12 ): 125 - 135 .

DING H Q , ZHANG P F , NIU H W H , et al . Cloud-based intelligent computing center ten-thousand card cluster innovation and practice [J ] . Telecommunications Science , 2024 , 40 ( 12 ): 125 - 135 .

YUAN B H , HE Y J , DAVIS J , et al . Decentralized training of foundation models in heterogeneous environments [C ] // Proceedings of the 36th Conference on Neural Information Processing Systems (NeurIPS 2022 ), New York : Curran Assiociates , 2024 : 25464 – 25477 .

SUN J , WANG D , QI B , et al . Decentralized training over 100 km based on optical transport network for artificial intelligence [C ] // Proceedings of the ECOC 2024 ; 50th European Conference on Optical Communication . Piscataway : IEEE Press , 2024 : 1459 - 1462 .

CAMERA M , OLSSON B E , BRUNO G . Beyond 100 Gbit/s: system implications towards 400G and 1 T[C ] // Proceedings of the 36th European Conference and Exhibition on Optical Communication . Piscataway : IEEE Press , 2010 : 1 - 15 .

WANG D , LI Y B , ZHANG D C , et al . Ultra-low-loss and large-effective-area fiber for 100 Gbit/s and beyond 100 Gbit/s coherent long-haul terrestrial transmission systems [J ] . Scientific Reports , 2019 , 9 ( 1 ): 17162 .

LORENCES-RIESGO A , BENDIMERAD D , LE-TRUNG K , et al . PCS-16QAM vs QPSK: what is the best choice for next-generation long-haul 400G? [C ] // Proceedings of the 2021 European Conference on Optical Communication (ECOC) . Piscataway : IEEE Press , 2021 : 1 - 4 .

ZUO M Q , YAN B L , GE D W , et al . 32- λ × 400 Gb/s single-carrier 120-GBaud QPSK coherent transmission over 3075-km G.652.D fiber link using OE-MCM prototype under field-deployed configuration [C ] // Proceedings of the 2023 Optical Fiber Communications Conference and Exhibition (OFC) . Piscataway : IEEE Press , 2023 : 1 - 3 .

ZUO M , WANG D , GE D , et al . Field trial of 40-λ×400-Gb/s single-carrier 138-GBd DP-QPSK real-time 6028 km transmission for computility network [C ] // Opto-Electronics and Communications Conference . Piscataway : IEEE press , 2024 : 1 - 3 .

GE D W , ZUO M Q , LIU H B , et al . Fully-loaded 80× 400 Gb/s DP-QPSK transmission with commercial 12-THz C6T L6T EDFAs over record distance of 7000 km [C ] // Proceedings of the 2023 Asia Communications and Photonics Conference/2023 International Photonics and Optoelectronics Meetings (ACP/POEM) . Piscataway : IEEE Press , 2023 : 1 - 4 .

ZUO M Q , ZHANG D C , GE D W , et al . Experimental demonstration of real-time fully-loaded $80 × 800 Gb/s transmission over 1253-km G . 652 .D fiber with commercial 12-THz C L-band EDFA and integrated WSS[C ] // Proceedings of the 2024 Asia Communications and Photonics Conference (ACP) and International Conference on Information Photonics and Optical Communications (IPOC) . Piscataway : IEEE Press , 2024: 1 - 3 .

中华人民共和国工业和信息化部 . 城域 N× 400 Gbit/s 光波分复用: YD/T 3964-2021 [S ] . 2021 .

Ministry of Industry and Information Technology of the People’s Republic of China . Technical requirements for metro N×400 Gbit/s optical wavelength division multiplexing (WDM) system: YD/T 3964-2021 [S ] . 2021 .

PUTTNAM B J , LUIS R S , PHILLIPS I , et al . 402 Tb/s GMI data-rate OESCLU-band transmission [C ] // Proceedings of the 2024 Optical Fiber Communications Conference and Exhibition (OFC) . Piscataway : IEEE Press , 2024 : 1 - 3 .

ZUO M , ZHANG D , WANG D , et al . First field demonstration of real-time sub-100-Tb/s transmission with net 1.2-Tb/s channels over 12-THz-wide super C+L band along 305-km G.652.D fiber [C ] // Proceedings of the 2025 Optical Fiber Communications Conference and Exhibition (OFC) . Piscataway : IEEE Press , 2025 : 1 - 3 .

ZHANG Y , ZUO M , QIU Q , et al . 214-Tb/s transmission over 2×75-km in the S+C+L band with >1-Tb/s/λ signals using only doped fiber amplifiers [C ] // Proceedings of the 2025 Optical Fiber Communications Conference and Exhibition (OFC) . Piscataway : IEEE Press , 2025 : 1 - 3 .

ZHANG A X , LIU Y Y , YAN B L , et al . Record real-time 128.7 Tbit/s DWDM transmission over 75km G.654.D fiber using S+C+L 17THz bandwidth lumped doped fiber amplifiers [C ] // Proceedings of the 2024 Asia Communications and Photonics Conference (ACP) and International Conference on Information Photonics and Optical Communications (IPOC) . Piscataway : IEEE Press , 2024 : 1 - 4 .

LI Y B , ZHANG D C , WANG Z W , et al . Field trial of concurrent co-cable and co-trench optical fiber online identification based on ensemble learning [J ] . Optics Express , 2023 , 31 ( 26 ): 42850 - 42865 .

LIU Y , WANG D , LI Y , et al . Highly-precise fiber co-route segment location with multi-modal vibration analysis and field demonstration for intelligent optical network [C ] // Proceedings of the 2025 Optical Fiber Communications Conference and Exhibition (OFC) . Piscataway : IEEE Press , 2025 : 1 - 3 .

连阳 . OTN 中保护倒换机制的研究 [J ] . 光通信技术 , 2009 , 33 ( 1 ): 16 - 18 .

LIAN Y . Research on protect switching of OTN [J ] . Optical Communication Technology , 2009 , 33 ( 1 ): 16 - 18 .

浏览量

813

下载量

CSCD

文章被引用时，请邮件提醒。

提交

工具集

关联资源

暂无数据