{"status":"ok","message-type":"work","message-version":"1.0.0","message":{"indexed":{"date-parts":[[2026,10,23]],"date-time":"2026-10-23T20:22:08Z","timestamp":1792786928730,"version":"4.5.0"},"reference-count":57,"publisher":"Institute of Electrical and Electronics Engineers (IEEE)","license":[{"start":{"date-parts":[[2026,1,1]],"date-time":"2026-01-01T00:00:00Z","timestamp":1767225600000},"content-version":"vor","delay-in-days":0,"URL":"https:\/\/ieeexplore.ieee.org\/Xplorehelp\/downloads\/license-information\/IEEE.html"},{"start":{"date-parts":[[2026,1,1]],"date-time":"2026-01-01T00:00:00Z","timestamp":1767225600000},"content-version":"stm-asf","delay-in-days":0,"URL":"https:\/\/doi.org\/10.15223\/policy-029"},{"start":{"date-parts":[[2026,1,1]],"date-time":"2026-01-01T00:00:00Z","timestamp":1767225600000},"content-version":"stm-asf","delay-in-days":0,"URL":"https:\/\/doi.org\/10.15223\/policy-037"},{"start":{"date-parts":[[2026,1,1]],"date-time":"2026-01-01T00:00:00Z","timestamp":1767225600000},"content-version":"tdm","delay-in-days":0,"URL":"https:\/\/ieeexplore.ieee.org\/Xplorehelp\/tdmrep.html"}],"funder":[{"name":"Hong Kong RGC TRS","award":["T41-603\/20R"],"award-info":[{"award-number":["T41-603\/20R"]}]},{"name":"TACC [57]"},{"name":"NSFC","award":["92582109"],"award-info":[{"award-number":["92582109"]}],"id":[{"id":"https:\/\/ror.org\/01h0zpd94","id-type":"ROR","asserted-by":"crossref"}]}],"content-domain":{"domain":[],"crossmark-restriction":false},"short-container-title":["IEEE Trans. Netw."],"published-print":{"date-parts":[[2026]]},"DOI":"10.1109\/ton.2026.3727576","type":"journal-article","created":{"date-parts":[[2026,8,26]],"date-time":"2026-08-26T19:05:38Z","timestamp":1787771138000},"page":"7424-7438","source":"Crossref","is-referenced-by-count":0,"title":["Understanding Communication Predictability of Distributed LLM Training"],"prefix":"10.1109","volume":"34","author":[{"ORCID":"https:\/\/orcid.org\/0000-0002-8228-2552","authenticated-orcid":false,"given":"Wenxue","family":"Li","sequence":"first","affiliation":[{"name":"The Hong Kong University of Science and Technology","place":["Hong Kong, SAR, China"]}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Xiangzhou","family":"Liu","sequence":"additional","affiliation":[{"name":"The Hong Kong University of Science and Technology","place":["Hong Kong, SAR, China"]}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Yuxuan","family":"Li","sequence":"additional","affiliation":[{"name":"The Hong Kong University of Science and Technology","place":["Hong Kong, SAR, China"]}],"role":[{"vocabulary":"crossref","role":"author"}]},{"ORCID":"https:\/\/orcid.org\/0000-0002-9502-7622","authenticated-orcid":false,"given":"Yilun","family":"Jin","sequence":"additional","affiliation":[{"name":"The Hong Kong University of Science and Technology","place":["Hong Kong, SAR, China"]}],"role":[{"vocabulary":"crossref","role":"author"}]},{"ORCID":"https:\/\/orcid.org\/0000-0002-3238-8500","authenticated-orcid":false,"given":"Han","family":"Tian","sequence":"additional","affiliation":[{"name":"The Hong Kong University of Science and Technology","place":["Hong Kong, SAR, China"]}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Zhenghang","family":"Ren","sequence":"additional","affiliation":[{"name":"The Hong Kong University of Science and Technology","place":["Hong Kong, SAR, China"]}],"role":[{"vocabulary":"crossref","role":"author"}]},{"ORCID":"https:\/\/orcid.org\/0000-0002-8380-1879","authenticated-orcid":false,"given":"Xudong","family":"Liao","sequence":"additional","affiliation":[{"name":"The Hong Kong University of Science and Technology","place":["Hong Kong, SAR, China"]}],"role":[{"vocabulary":"crossref","role":"author"}]},{"ORCID":"https:\/\/orcid.org\/0000-0003-3131-374X","authenticated-orcid":false,"given":"Zhizhen","family":"Zhong","sequence":"additional","affiliation":[{"name":"Massachusetts Institute of Technology","place":["Cambridge, MA, USA"]}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Guyue","family":"Liu","sequence":"additional","affiliation":[{"name":"Peking University","place":["Beijing, China"]}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Ying","family":"Zhang","sequence":"additional","affiliation":[{"name":"Meta","place":["Menlo Park, CA, USA"]}],"role":[{"vocabulary":"crossref","role":"author"}]},{"ORCID":"https:\/\/orcid.org\/0000-0003-2587-6028","authenticated-orcid":false,"given":"Kai","family":"Chen","sequence":"additional","affiliation":[{"name":"The Hong Kong University of Science and Technology","place":["Hong Kong, SAR, China"]}],"role":[{"vocabulary":"crossref","role":"author"}]}],"member":"263","reference":[{"key":"ref1","first-page":"1877","article-title":"Language models are few-shot learners","volume-title":"Proc. Adv. Neural Inf. Process. Syst.","author":"Brown"},{"key":"ref2","article-title":"LLaMA: Open and efficient foundation language models","author":"Touvron","year":"2023","journal-title":"arXiv:2302.13971"},{"key":"ref3","article-title":"A survey of large language models","author":"Xin Zhao","year":"2023","journal-title":"arXiv:2303.18223"},{"key":"ref4","article-title":"Domain-specific communication optimization for distributed DNN training","author":"Wang","year":"2020","journal-title":"arXiv:2008.08445"},{"key":"ref5","first-page":"1141","article-title":"Accelerating neural recommendation training with embedding scheduling","volume-title":"Proc. NSDI","author":"Zeng"},{"key":"ref6","article-title":"Characterization of large language model development in the datacenter","author":"Hu","year":"2024","journal-title":"arXiv:2403.07648"},{"key":"ref7","first-page":"745","article-title":"MegaScale: Scaling large language model training to more than 10,000 GPUs","volume-title":"Proc. NSDI","author":"Jiang"},{"key":"ref8","doi-asserted-by":"publisher","DOI":"10.1145\/3341301.3359646"},{"key":"ref9","first-page":"1","article-title":"Efficient large-scale language model training on GPU clusters using megatron-LM","volume-title":"Proc. SC21: Int. Conf. High Perform. Comput., Netw., Storage Anal.","author":"Narayanan"},{"key":"ref10","first-page":"1191","article-title":"THC: Accelerating distributed deep learning using tensor homomorphic compression","volume-title":"Proc. NSDI","author":"Li"},{"key":"ref11","doi-asserted-by":"publisher","DOI":"10.1145\/3477132.3483553"},{"key":"ref12","doi-asserted-by":"publisher","DOI":"10.1145\/3552326.3567505"},{"key":"ref13","doi-asserted-by":"publisher","DOI":"10.1145\/3452296.3472904"},{"key":"ref14","doi-asserted-by":"publisher","DOI":"10.1145\/3341301.3359642"},{"key":"ref15","doi-asserted-by":"publisher","DOI":"10.1109\/INFOCOM48880.2022.9796752"},{"key":"ref16","first-page":"132","article-title":"Priority-based parameter propagation for distributed DNN training","volume-title":"Proc. Mach. Learn. Syst.","author":"Jayarajan"},{"key":"ref17","doi-asserted-by":"publisher","DOI":"10.1145\/1672308.1672325"},{"key":"ref18","doi-asserted-by":"publisher","DOI":"10.1145\/2785956.2787472"},{"key":"ref19","doi-asserted-by":"publisher","DOI":"10.1145\/1879141.1879175"},{"key":"ref20","first-page":"595","article-title":"Gandiva: Introspective cluster scheduling for deep learning","volume-title":"Proc. OSDI","author":"Xiao"},{"key":"ref21","article-title":"CASSINI: Network-aware job scheduling in machine learning clusters","author":"Rajasekaran","year":"2023","journal-title":"arXiv:2308.00852"},{"key":"ref22","doi-asserted-by":"publisher","DOI":"10.1145\/3297858.3304072"},{"key":"ref23","first-page":"739","article-title":"TopoOpt: Co-optimizing network topology and parallelization strategy for distributed training jobs","volume-title":"Proc. NSDI","author":"Wang"},{"key":"ref24","article-title":"How to build low-cost networks for large language models (without sacrificing performance)?","author":"Wang","year":"2023","journal-title":"arXiv:2307.12169"},{"key":"ref25","doi-asserted-by":"publisher","DOI":"10.1145\/3485447.3511981"},{"key":"ref26","first-page":"337","article-title":"Daydream: Accurately estimating the efficacy of optimizations for DNN training","volume-title":"Proc. USENIX Annu. Tech. Conf. (USENIX ATC 20)","author":"Zhu"},{"key":"ref27","doi-asserted-by":"publisher","DOI":"10.1145\/3405671.3405810"},{"key":"ref28","first-page":"559","article-title":"Alpa: Automating inter-and intra-operator parallelism for distributed deep learning","volume-title":"Proc. OSDI","author":"Zheng"},{"key":"ref29","first-page":"1","article-title":"Beyond data and model parallelism for deep neural networks","volume-title":"Proc. Mach. Learn. Syst.","author":"Jia"},{"key":"ref30","doi-asserted-by":"publisher","DOI":"10.1145\/3663408.3663409"},{"key":"ref31","article-title":"Tictac: Accelerating distributed deep learning with communication scheduling","author":"Hashemi","year":"2019","journal-title":"Proc. Mach. Learn. Syst."},{"key":"ref32","first-page":"485","article-title":"Tiresias: A GPU cluster manager for distributed deep learning","volume-title":"Proc. NSDI","author":"Gu"},{"key":"ref33","doi-asserted-by":"publisher","DOI":"10.1145\/3651890.3672239"},{"key":"ref34","doi-asserted-by":"publisher","DOI":"10.1145\/3651890.3672233"},{"key":"ref35","doi-asserted-by":"publisher","DOI":"10.1145\/3651890.3672265"},{"key":"ref36","first-page":"473","article-title":"Accelerating design space exploration for LLM training systems with multi-experiment parallel simulation","volume-title":"Proc. NSDI","author":"Gui"},{"key":"ref37","doi-asserted-by":"publisher","DOI":"10.1145\/3603269.3610851"},{"key":"ref38","volume-title":"Ns-3, a Discrete-Event Network Simulator for Internet Systems","year":"2026"},{"key":"ref39","first-page":"19","article-title":"Communication efficient distributed machine learning with the parameter server","volume-title":"Proc. Adv. Neural Inf. Process. Syst.","author":"Li"},{"key":"ref40","first-page":"463","article-title":"A unified architecture for accelerating distributed DNN training in heterogeneous GPU\/CPU clusters","volume-title":"Proc. OSDI","author":"Jiang"},{"key":"ref41","doi-asserted-by":"publisher","DOI":"10.1177\/1094342005051521"},{"key":"ref42","volume-title":"NCCL","year":"2024"},{"key":"ref43","doi-asserted-by":"publisher","DOI":"10.1147\/JRD.2019.2947013"},{"key":"ref44","doi-asserted-by":"publisher","DOI":"10.48550\/arxiv.1811.06965"},{"key":"ref45","first-page":"341","article-title":"Reducing activation recomputation in large transformer models","volume-title":"Proc. Mach. Learn. Syst.","author":"Korthikanti"},{"issue":"120","key":"ref46","first-page":"1","article-title":"Switch transformers: Scaling to trillion parameter models with simple and efficient sparsity","volume":"23","author":"Fedus","year":"2022","journal-title":"J. Mach. Learn. Res."},{"key":"ref47","article-title":"Outrageously large neural networks: The sparsely-gated mixture-of-experts layer","author":"Shazeer","year":"2017","journal-title":"arXiv:1701.06538"},{"key":"ref48","doi-asserted-by":"publisher","DOI":"10.18653\/v1\/2024.acl-long.70"},{"key":"ref49","article-title":"Qwen2 technical report","volume-title":"arXiv:2407.10671","author":"Yang","year":"2024"},{"key":"ref50","first-page":"18332","article-title":"DeepSpeed-MoE: Advancing mixture-of-experts inference and training to power next-generation AI scale","volume-title":"Proc. Int. Conf. Mach. Learn.","author":"Rajbhandari"},{"key":"ref51","article-title":"GShard: Scaling giant models with conditional computation and automatic sharding","author":"Lepikhin","year":"2020","journal-title":"arXiv:2006.16668"},{"key":"ref52","volume-title":"Megatron-DeepSpeed","year":"2021"},{"key":"ref53","article-title":"Language models are unsupervised multitask learners","author":"Radford","year":"2019","journal-title":"OpenAI Blog"},{"key":"ref54","volume-title":"N. D. S. Infrastructure","year":"2024"},{"key":"ref55","article-title":"An empirical model of large-batch training","author":"McCandlish","year":"2018","journal-title":"arXiv:1812.06162"},{"key":"ref56","doi-asserted-by":"publisher","DOI":"10.1109\/SC41405.2020.00024"},{"key":"ref57","doi-asserted-by":"publisher","DOI":"10.1145\/3669940.3707266"}],"container-title":["IEEE Transactions on Networking"],"original-title":[],"link":[{"URL":"http:\/\/xplorestaging.ieee.org\/ielx8\/10723154\/11317935\/11667689.pdf?arnumber=11667689","content-type":"unspecified","content-version":"vor","intended-application":"similarity-checking"}],"deposited":{"date-parts":[[2026,10,13]],"date-time":"2026-10-13T07:00:46Z","timestamp":1791874846000},"score":1,"resource":{"primary":{"URL":"https:\/\/ieeexplore.ieee.org\/document\/11667689\/"}},"subtitle":[],"short-title":[],"issued":{"date-parts":[[2026]]},"references-count":57,"URL":"https:\/\/doi.org\/10.1109\/ton.2026.3727576","relation":{},"ISSN":["2998-4157"],"issn-type":[{"value":"2998-4157","type":"electronic"}],"subject":[],"published":{"date-parts":[[2026]]}}}