Towards Network Infrastructure Research for the Era of Large Language Models: Challenges, Practices, and Prospects

Zhai Ennan; Cao Jiamin; Qian Kun; Guan Yu

doi:10.7544/issn1000-1239.202440576

Journal of Computer Research and Development > 2024 > 61(11): 2664-2677. > DOI: 10.7544/issn1000-1239.202440576 CSTR: 32373.14.issn1000-1239.202440576

Zhai Ennan, Cao Jiamin, Qian Kun, Guan Yu. Towards Network Infrastructure Research for the Era of Large Language Models: Challenges, Practices, and Prospects[J]. Journal of Computer Research and Development, 2024, 61(11): 2664-2677. DOI: 10.7544/issn1000-1239.202440576

Citation:

PDF (1656 KB)

Towards Network Infrastructure Research for the Era of Large Language Models: Challenges, Practices, and Prospects

Alibaba Cloud Computing Ltd., Hangzhou 310000

More Information

Author Bio:
Zhai Ennan: born in 1984. PhD. Member of CCF. His main research interests include computer network and distributed systems

Cao Jiamin: born in 1997. PhD. Member of CCF. Her main research interests include AI infrastructure and programmable networks

Qian Kun: born in 1993. PhD. Member of CCF. His main research interests include AI infrastructure and storage

Guan Yu: born in 1993. PhD. His main research interests include AI infrastructure and video streaming
Received Date: June 24, 2024
Revised Date: September 19, 2024
Available Online: September 26, 2024

Graphical Abstract

Abstract

Abstract

Large language models (LLMs) with hundreds of billions of parameters have brought significant technological and business transformations to today’s AI and cloud services. However, there exists a fundamental difference in network pattern between LLM training and general cloud computing (e.g., Amazon EC2 Elastic compute service), leading to a variety of new challenges. These challenges mainly include load balancing difficulties due to the traffic pattern difference (Challenge 1), the impact of multi-job communication contention on GPU utilization (Challenge 2), and high sensitivity to network failures (Challenge 3). Therefore, data center network technologies designed for general cloud computing (e.g., network architecture, routing, communication scheduling, and reliability) are no longer suitable for LLM training today. This necessitates the development of new data center networks and accompanying technical solutions specifically for LLM training. We introduce Alibaba Cloud’s high-performance network (HPN) and the multi-job communication scheduling approach Crux, designed to address the aforementioned challenges. HPN introduces a two-layer, dual-plane network architecture, which not only achieves high-speed interconnectivity for 15 000 GPUs within a Pod but also ensures precise routing suitable for LLM training (addressing Challenge 1). Furthermore, HPN proposes a novel dual-top-of-rack (ToR) design, replacing the traditional single ToR switch connection in data center networks and fundamentally avoiding single-point failure reliability risks (partially addressing Challenge 3). To tackle Challenge 2, Crux reduces the NP-complete problem of optimizing GPU utilization by modeling it as a communication scheduling issue related to GPU computational intensity. Crux then proposes an algorithm that prioritizes the flows of job with higher GPU computational intensity, significantly reducing multi-job communication contention and improving GPU utilization. Compared with the state-of-the-art efforts, Crux increases GPU utilization by up to 23%. Both HPN and Crux have been deployed and used in Alibaba Cloud production for over eight months and will continue to evolve and iterate. Building on this, we further envision possible research directions in the field of LLM training and inference, providing guidance for subsequent work.
- AI infrastructure,
- large language model (LLMs),
- large models,
- model training,
- data center networks,
- collective communication,
- communication scheduling

FullText(HTML)

References (34)

References

[1]	OpenAI, Josh A, Adler S, et al. GPT−4 technical report. [J]. arXiv preprint, arXiv: 2303.08774, 2024
[2]	OpenAI. Introducing ChatGPT[EB/OL]. [2022-11-30]. https://openai.com/blog/chatgpt
[3]	MLSYS ORG. Vicuna: An open-source chatbot impressing GPT−4 with 90% ChatGPT quality[EB/OL]. [2023-03-30]. https://lmsys.org/blog/2023-03-30-vicuna/
[4]	NVIDIA. Megatron-LM. [EB/OL]. [2024-06-19]. https://github.com/NVIDIA/‌Megatron-LM
[5]	Microsoft Research. DeepSpeed[EB/OL]. [2024-06-19]. https://www.microsoft.com/en-us/‌research/project/deepspeed/
[6]	Rajasekaran S, Ghobadi M, Akella A. CASSINI: Network-aware job scheduling in machine learning clusters[C]//Proc of USENIX NSDI. Berkeley, CA: USENIX Association, 2024: 1403−1420
[7]	Jiang Ziheng, Lin Haibin, Zhong Yinmin, et al. MegaScale: Scaling large language model training to more than 10, 000 GPUs[C]//Proc of USENIX NSDI. Berkeley, CA: USENIX Association, 2024: 745−760
[8]	Qian Kun, Xi Yongqing, Cao Jiamin, et al. Alibaba HPN: A data center network for large language model training[C]//Proc of ACM SIGCOMM. New York: ACM, 2024: 691−706
[9]	Cao Jiamin, Guan Yu, Qian Kun, et al. Crux: GPU-efficient communication scheduling for deep learning training[C]//Proc of ACM SIGCOMM. New York: ACM, 2024: 1−15
[10]	Microsoft DeepSpeed. Model checkpointing[EB/OL]. [2023-01-31]. https://deepspeed.readthedocs.‌io/en/latest/model-checkpointing.html
[11]	Alizadeh M, Edsall T, Dharmapurikar S, et al. CONGA: Distributed congestion-aware load balancing for datacenters[C]//Proc of ACM SIGCOMM. New York: ACM, 2014: 503–514
[12]	Dixit A, Prakash P, Hu Y C, et al. On the impact of packet spraying in data center networks[C]//Proc of IEEE INFOCOM. Piscataway, NJ: IEEE, 2013: 2130–2138
[13]	Ghorbani S, Yang Zibin, Godfrey P B, et al. DRILL: Micro load balancing for low-latency data center networks[C]//Proc of ACM SIGCOMM. New York: ACM, 2017: 225–238
[14]	Katta N, Hira M, Ghag A, et al. CLOVE: How I learned to stop worrying about the core and love the edge[C]//Proc of ACM Workshop on Hot Topics in Networks. New York: ACM, 2016: 155–161
[15]	Katta N, Hira M, Kim C, et al. HULA: Scalable load balancing using programmable data planes[C]//Proc of ACM SOSR. New York: ACM, 2016: 1–12
[16]	Qureshi M A, Cheng Yuchung, Yin Qianwen, et al. PLB: Congestion signals are simple and effective for network load balancing[C]//Proc of ACM SIGCOMM. New York: ACM, 2022: 207–218
[17]	Sen S, Shue D, Ihm S, et al. Scalable, optimal flow routing in datacenters via local link balancing[C]//Proc of ACM CoNEXT. New York: ACM, 2013: 151–162
[18]	Vanini E, Pan Rong, Alizadeh M, et al. Let it flow: Resilient asymmetric load balancing with flowlet switching[C]//Proc of USENIX NSDI. Berkeley, CA: USENIX Association, 2017: 407–420
[19]	Zats D, Das T, Mohan P, et al. DeTail: Reducing the flow completion time tail in datacenter networks[C]//Proc of ACM SIGCOMM. New York: ACM, 2012: 139–150
[20]	Zhang Hong, Zhang Junxue, Bai Wei, et al. Resilient datacenter load balancing in the wild[C]//Proc of ACM SIGCOMM. New York: ACM, 2017: 253–266
[21]	Agarwal S, Rajakrishnan S, Narayan A, et al. Sincronia: Near-optimal network design for coflows[C]//Proc of ACM SIGCOMM. New York: ACM, 2018: 16−29
[22]	Shah A, Chidambaram V, Cowan M, et al. TACCL: Guiding collective algorithm synthesis using communication sketches[C]//Proc of USENIX NSDI. Berkeley, CA: USENIX Association, 2023: 593−612
[23]	NVIDIA. NVLink and NVSwitch[EB/OL]. [2024-06-19]. https://www.nvidia.com/‌en-us/data-center/nvlink/
[24]	Meta. Meta’s evolution of network for AI[EB/OL]. [2023-11-01]. https://‌www.youtube.com/watch?v=5gOOtFySrqA
[25]	NVIDIA. NVIDIA DGX SuperPOD: Next generation scalable infrastructure for AI leadership[EB/OL]. [2023-09-22]. https: /docs. nvidia. com/dgx-superpod-‌reference-architecture-dgx-h100. pdf
[26]	Al-Fares M, Loukissas A, Vahdat A. A scalable, commodity data center network architecture[C]//Proc of ACM SIGCOMM. New York: ACM, 2008: 63–74
[27]	Greenberg A, Hamilton J R, Jain N, et al. VL2: A scalable and flexible data center network[C]//Proc of ACM SIGCOMM. New York: ACM, 2009: 51–62
[28]	Bai Wei, Abdeen S S, Agrawal A, et al. Empowering Azure storage with RDMA[C]//Proc of USENIX NSDI. Berkeley, CA: USENIX Association, 2023: 49–67
[29]	Poutievski L, Mashayekhi O, Ong J, et al. Jupiter evolving: Transforming Google’s datacenter network via optical circuit switches and software-defined networking[C]//Proc of ACM SIGCOMM. New York: ACM, 2022: 66–85
[30]	Linux GNU. Linux bonding modes[EB/OL]. [2010-01-08]. https://thelinuxcluster.‌com/2010/01/08/linux-bonding-modes/
[31]	IEEE Standard SA. IEEE 802.3ad[EB/OL]. [2000-06-28]. https://standards.ieee.org/‌ieee/802.3ad/1088/
[32]	IEEE Standard SA. IEEE standard for information technology – local and metropolitan area networks – specific requirements – part 3: CSMA/CD access method and physical layer specifications amendment 5: Media access control parameters, physical layers, and management parameters for energy-efficient Ethernet[EB/OL]. [2010-10-27]. https://standards.ieee.org/‌ieee/802.3az/4270/
[33]	Zhang Zhehui, Zheng Haiyang, Hu Jiayao, et al. Hashing linearity enables relative path control in data centers[C]//Proc of USENIX ATC. Berkeley, CA: USENIX Association, 2021: 855–862
[34]	NVIDIA. InfiniBand networking solutions[EB/OL]. [2024-06-11]. https://www.nvidia.com/en-us/networking/products/infiniband/

[1]	Li Nan, Ding Yidong, Jiang Haoyu, Niu Jiafei, Yi Ping. Jailbreak Attack for Large Language Models: A Survey[J]. Journal of Computer Research and Development, 2024, 61(5): 1156-1181. DOI: 10.7544/issn1000-1239.202330962
[2]	Wang Mengru, Yao Yunzhi, Xi Zekun, Zhang Jintian, Wang Peng, Xu Ziwen, Zhang Ningyu. Safety Analysis of Large Model Content Generation Based on Knowledge Editing[J]. Journal of Computer Research and Development, 2024, 61(5): 1143-1155. DOI: 10.7544/issn1000-1239.202330965
[3]	Chen Xuanting, Ye Junjie, Zu Can, Xu Nuo, Gui Tao, Zhang Qi. Robustness of GPT Large Language Models on Natural Language Processing Tasks[J]. Journal of Computer Research and Development, 2024, 61(5): 1128-1142. DOI: 10.7544/issn1000-1239.202330801
[4]	Chen Huimin, Liu Zhiyuan, Sun Maosong. The Social Opportunities and Challenges in the Era of Large Language Models[J]. Journal of Computer Research and Development, 2024, 61(5): 1094-1103. DOI: 10.7544/issn1000-1239.202330700
[5]	Yang Yi, Li Ying, Chen Kai. Vulnerability Detection Methods Based on Natural Language Processing[J]. Journal of Computer Research and Development, 2022, 59(12): 2649-2666. DOI: 10.7544/issn1000-1239.20210627
[6]	Pan Xuan, Xu Sihan, Cai Xiangrui, Wen Yanlong, Yuan Xiaojie. Survey on Deep Learning Based Natural Language Interface to Database[J]. Journal of Computer Research and Development, 2021, 58(9): 1925-1950. DOI: 10.7544/issn1000-1239.2021.20200209
[7]	Zheng Haibin, Chen Jinyin, Zhang Yan, Zhang Xuhong, Ge Chunpeng, Liu Zhe, Ouyang Yike, Ji Shouling. Survey of Adversarial Attack, Defense and Robustness Analysis for Natural Language Processing[J]. Journal of Computer Research and Development, 2021, 58(8): 1727-1750. DOI: 10.7544/issn1000-1239.2021.20210304
[8]	Pan Xudong, Zhang Mi, Yan Yifan, Lu Yifan, Yang Min. Evaluating Privacy Risks of Deep Learning Based General-Purpose Language Models[J]. Journal of Computer Research and Development, 2021, 58(5): 1092-1105. DOI: 10.7544/issn1000-1239.2021.20200908
[9]	Bao Yang, Yang Zhibin, Yang Yongqiang, Xie Jian, Zhou Yong, Yue Tao, Huang Zhiqiu, Guo Peng. An Automated Approach to Generate SysML Models from Restricted Natural Language Requirements in Chinese[J]. Journal of Computer Research and Development, 2021, 58(4): 706-730. DOI: 10.7544/issn1000-1239.2021.20200757
[10]	Che Haiyan, Feng Tie, Zhang Jiachen, Chen Wei, and Li Dali. Automatic Knowledge Extraction from Chinese Natural Language Documents[J]. Journal of Computer Research and Development, 2013, 50(4): 834-842.

Supplements (1)

Supplements
Article Video
- Video Player is loading.
  Current Time 0:00
  Duration -:-
  Loaded: 0%
  Stream Type LIVE
  Remaining Time -:-
  
  1x
  Chapters
  descriptions off, selected
  captions settings, opens captions settings dialog
  captions off, selected
  This is a modal window.
  The media could not be loaded, either because the server or network failed or because the format is not supported.
  Beginning of dialog window. Escape will cancel and close the window.
  TextColorTransparency
  BackgroundColorTransparency
  WindowColorTransparency
  Font Size
  Text Edge Style
  Font Family
  End of dialog window.
  Video Abstract
Other Related Supplements
- Video
  https://crad.ict.ac.cn/news/537

Cited By

Cited by

Periodical cited type(66)

1.	袁良志，海佳丽，汪润，邓文萍，肖勇，常凯. 知识图谱驱动的中医药标准数字化探索与实践. 中医药导报. 2025(01): 225-230 .
2.	范定容，王倩倩，沈奥，彭露. 从ChatGPT到Sora：人工智能在医学教育中的应用潜力与挑战. 中国医学教育技术. 2025(01): 33-40 .
3.	刘园园，王银刚. ChatGPT影响大学生判断能力：双向机理与对策. 湖北成人教育学院学报. 2025(01): 29-34 .
4.	魏昱，刘卫. 人工智能生成内容在服装设计中的应用现状. 毛纺科技. 2025(01): 134-142 .
5.	李冰，鲜勇，雷刚，苏娟. ChatGPT架构下课程智能教学助手建设探讨. 教育教学论坛. 2025(03): 45-48 .
6.	梁炜，许振宇. 大语言模型赋能舆情治理现代化：价值、风险与路径. 中国应急管理科学. 2025(01): 93-103 .
7.	刘邦奇，聂小林，王士进，袁婷婷，朱洪军，赵子琪，朱广袤. 生成式人工智能与未来教育形态重塑：技术框架、能力特征及应用趋势. 电化教育研究. 2024(01): 13-20 .
8.	秦涛，杜尚恒，常元元，王晨旭. ChatGPT的工作原理、关键技术及未来发展趋势. 西安交通大学学报. 2024(01): 1-12 .
9.	张小朝. AIGC在商旅行业中的应用探索. 广东通信技术. 2024(01): 75-79 .
10.	廉霄兴，宋勇，朱军，王淑玲，叶晓舟，欧阳晔. 基于双通道理论的通信认知增强技术研究. 电信科学. 2024(01): 123-135 .
11.	杨永恒. 人工智能时代社会科学研究的“变”与“不变”. 人民论坛·学术前沿. 2024(04): 96-105 .
12.	刘英祥，张琳. 生成式人工智能技术在海事管理工作中的应用探索. 航海. 2024(02): 62-64 .
13.	吕静，何平，王永芬，冉朝霞，曹钦兴，古文帆，彭敏，田敏. ChatGPT在医学领域研究态势的文献计量学分析. 医学与哲学. 2024(07): 30-35 .
14.	王益君，董韵美. 公众对人工智能的认知与情感态度——以ChatGPT为例. 知识管理论坛. 2024(01): 16-29 .
15.	陈雷. ChatGPT在公安院校教育教学中的应用及影响. 太原城市职业技术学院学报. 2024(02): 85-88 .
16.	尤冲，李彦兵. 基于ChatGPT大语言模型应用的公共体育服务智能化:指征、风险及其规制. 南京体育学院学报. 2024(02): 1-12 .
17.	杨胜钦. 从ChatGPT看AI对电信网络诈骗犯罪治理的影响. 犯罪与改造研究. 2024(05): 26-33 .
18.	王春英，姚亚妮，滕白莹. 生成式人工智能嵌入敏捷政府建设：影响、风险与应对. 北京行政学院学报. 2024(03): 73-83 .
19.	王雯，李永智. 国际生成式人工智能教育应用与省思. 开放教育研究. 2024(03): 37-44 .
20.	张智义. 体认语言学视阈下ChatGPT语言生成及性能研究. 外语研究. 2024(03): 20-25+43+112 .
21.	余淑珍，单俊豪，闫寒冰. 情感计算赋能个性化教学：逻辑框架、问题解构与多元重塑. 现代远距离教育. 2024(02): 53-61 .
22.	高尚. 大语言模型与中台：共融还是替代？. 科技与金融. 2024(05): 59-62 .
23.	郭亚军，马慧芳，张鑫迪，冯思倩. ChatGPT赋能图书馆知识服务：原理、场景与进路. 图书馆建设. 2024(03): 60-68 .
24.	高雪松，黄蕴华，王斌. 基于专利数据的生成式人工智能技术栈创新态势研究. 东北财经大学学报. 2024(04): 53-61 .
25.	张渊. ChatGPT文本的生成机制与文本特性分析. 重庆文理学院学报(社会科学版). 2024(04): 105-114 .
26.	罗仕鉴，于慧伶，易珮琦. 数智时代工业设计知识生产新范式. 机械设计. 2024(08): 6-10 .
27.	徐炳文. 基于ChatGPT的人工智能交互技术工业物联网平台研究. 工业控制计算机. 2024(08): 132-134 .
28.	Deyi Li，Jialun Yin，Tianlei Zhang，Wei Han，Hong Bao. The Four Most Basic Elements In Machine Cognition. Data Intelligence. 2024(02): 297-319 .
29.	黄语，刘海洋，常海军，杨远松. 基于ChatGPT工作模式的AI工具在BIM技术中的潜在应用与实现途径. 科技创新与应用. 2024(26): 181-184+188 .
30.	李琳娜，丁楷，韩红旗，王力，李艾丹. 基于知识图谱的中文科技文献问答系统构建研究. 中国科技资源导刊. 2024(04): 51-62 .
31.	裴炳森，李欣，蒋章涛，刘明帅. 基于大语言模型的公安专业小样本知识抽取方法研究. 计算机科学与探索. 2024(10): 2630-2642 .
32.	李克寒，余丽媛，邵企能，蒋可，乌丹旦. 大语言模型在口腔住院医师规范化培训中的应用构想. 中国卫生产业. 2024(07): 155-158 .
33.	钟厚涛. 生成式人工智能给翻译实践带来的机遇与挑战. 北京翻译. 2024(00): 238-250 .
34.	张夏恒，马妍. AIGC在应急情报服务中的应用研究. 图书馆工作与研究. 2024(11): 60-67 .
35.	崔金满，李冬梅，田萱，孟湘皓，杨宇，崔晓晖. 提示学习研究综述. 计算机工程与应用. 2024(23): 1-27 .
36.	周代数，魏杉汀. 人工智能驱动的科学研究第五范式：演进、机制与影响. 中国科技论坛. 2024(12): 97-107 .
37.	钱力，张智雄，伍大勇，常志军，于倩倩，胡懋地，刘熠. 科技文献大模型:方法、框架与应用. 中国图书馆学报. 2024(06): 45-58 .
38.	潘崇佩，廖康启，孔勇发. 生成式人工智能背景下的近代物理实验教学改革. 实验室研究与探索. 2024(12): 117-122 .
39.	李德毅，刘玉超，殷嘉伦. 认知机器如何创造. 中国基础科学. 2024(06): 1-11 .
40.	李德毅，张天雷，韩威，海丹，鲍泓，高洪波. 认知机器的结构和激活. 智能系统学报. 2024(06): 1604-1613 .
41.	蔡昌，庞思诚. ChatGPT的智能性及其在财税领域的应用. 商业会计. 2023(09): 41-46 .
42.	于书娟，卢小雪，赵磊磊. 教育人工智能变革的基本逻辑与发展进路. 当代教育科学. 2023(05): 40-49 .
43.	曹克亮. ChatGPT：意识形态家的机器学转向及后果. 统一战线学研究. 2023(04): 134-144 .
44.	宋恺，屈蕾蕾，杨萌科. 生成式人工智能的治理策略研究. 信息通信技术与政策. 2023(07): 83-88 .
45.	陈凌云，姚宽达，王茜，方安，李刚. ChatGPT:研究进展、模型创新及医学信息研究应用场景优化. 医学信息学杂志. 2023(07): 18-23+29 .
46.	彭强，李羿卫. 自然用户界面在智能家居系统中的应用路径创新研究：生成式人工智能技术的调节作用. 包装工程. 2023(16): 454-463 .
47.	杨军农，王少波. 类ChatGPT技术嵌入政务服务网的应用场景、风险隐患与实施建议. 信息与电脑(理论版). 2023(10): 183-186 .
48.	政光景，吕鹏. 生成式人工智能与哲学社会科学新范式的涌现. 江海学刊. 2023(04): 132-142+256 .
49.	吴梦妮. 社交媒体传播视域下玩具企业应用AI技术实施营销的实践路径. 玩具世界. 2023(04): 144-146 .
50.	李德毅，殷嘉伦，张天雷，韩威，鲍泓. 机器认知四要素说. 中国基础科学. 2023(03): 1-10+22 .
51.	王洁. ChatGPT对知识服务的五大变革. 图书馆. 2023(09): 10-16 .
52.	刘乃嘉. 基于ChatGPT的矿山工程风险评估预警系统实现探讨. 企业科技与发展. 2023(08): 44-47 .
53.	裴炳森，李欣，吴越. 基于ChatGPT的电信诈骗案件类型影响力评估. 计算机科学与探索. 2023(10): 2413-2425 .
54.	张新新，丁靖佳. 生成式智能出版的技术原理与流程革新. 图书情报知识. 2023(05): 68-76 .
55.	张新新，黄如花. 生成式智能出版的应用场景、风险挑战与调治路径. 图书情报知识. 2023(05): 77-86+27 .
56.	陈靖. ChatGPT的类人想象与安全风险分析. 网络空间安全. 2023(04): 8-12 .
57.	李佩芳，陈佳丽，宁宁，王立群，张涵旎. ChatGPT在医学领域的应用进展及思考. 华西医学. 2023(10): 1456-1460 .
58.	朱敏锐，郜云帆，黄勇. 以新时代优良学风涵养新时代外语人才. 北京教育(高教). 2023(11): 35-37 .
59.	丁红菊. 消解与重构：人工智能技术对新闻业的影响——基于对ChatGPT的研究. 运城学院学报. 2023(05): 57-62 .
60.	李钥，淮盼盼，杨辉. ChatGPT在护理教育中的应用状况及优劣分析. 护理学杂志. 2023(21): 117-121 .
61.	张绍龙. 基于ChatGPT的人工智能技术应用. 集成电路应用. 2023(11): 200-201 .
62.	崔克克，孙冲，李辉，赵凌飞. 浅谈水泥企业数字化转型发展. 中国水泥. 2023(12): 28-33 .
63.	单琳，王文娟，刘舒萌. ChatGPT在医学分子生物学教学中的应用. 基础医学教育. 2023(12): 1084-1086 .
64.	李德毅，刘玉超，任璐. 人工智能看智慧. 科学与社会. 2023(04): 131-149 .
65.	付翔，魏晓伟，张浩，徐宁. 数字安全角度下审视和剖析ChatGPT. 航空兵器. 2023(06): 117-122 .
66.	黄婷，刘力凯. 基于大模型的数智化语言教学探索与应用. 连云港职业技术学院学报. 2023(04): 73-79 .