{"status":"ok","message-type":"work","message-version":"1.0.0","message":{"indexed":{"date-parts":[[2026,7,15]],"date-time":"2026-07-15T17:02:59Z","timestamp":1784134979263,"version":"3.55.0"},"reference-count":142,"publisher":"Institute of Electrical and Electronics Engineers (IEEE)","issue":"12","license":[{"start":{"date-parts":[[2024,12,1]],"date-time":"2024-12-01T00:00:00Z","timestamp":1733011200000},"content-version":"vor","delay-in-days":0,"URL":"https:\/\/creativecommons.org\/licenses\/by-nc-nd\/4.0\/"}],"funder":[{"name":"National Key Research and Development Program of China","award":["2021ZD0113000"],"award-info":[{"award-number":["2021ZD0113000"]}]}],"content-domain":{"domain":[],"crossmark-restriction":false},"short-container-title":["IEEE Trans. Parallel Distrib. Syst."],"published-print":{"date-parts":[[2024,12]]},"DOI":"10.1109\/tpds.2024.3476390","type":"journal-article","created":{"date-parts":[[2024,10,7]],"date-time":"2024-10-07T17:51:20Z","timestamp":1728323480000},"page":"2463-2478","source":"Crossref","is-referenced-by-count":9,"title":["A Survey on Performance Modeling and Prediction for Distributed DNN Training"],"prefix":"10.1109","volume":"35","author":[{"given":"Zhenhua","family":"Guo","sequence":"first","affiliation":[{"name":"IEIT SYSTEMS Company, Ltd., Jinan, China"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"ORCID":"https:\/\/orcid.org\/0000-0002-6029-3744","authenticated-orcid":false,"given":"Yinan","family":"Tang","sequence":"additional","affiliation":[{"name":"IEIT SYSTEMS Company, Ltd., Jinan, China"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"ORCID":"https:\/\/orcid.org\/0000-0002-7656-6428","authenticated-orcid":false,"given":"Jidong","family":"Zhai","sequence":"additional","affiliation":[{"name":"Tsinghua University, Beijing, China"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"ORCID":"https:\/\/orcid.org\/0000-0002-8224-9891","authenticated-orcid":false,"given":"Tongtong","family":"Yuan","sequence":"additional","affiliation":[{"name":"Beijing University of Technology, Beijing, China"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"ORCID":"https:\/\/orcid.org\/0000-0002-4128-966X","authenticated-orcid":false,"given":"Jian","family":"Jin","sequence":"additional","affiliation":[{"name":"CAICT, Beijing, China"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Li","family":"Wang","sequence":"additional","affiliation":[{"name":"IEIT SYSTEMS Company, Ltd., Jinan, China"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Yaqian","family":"Zhao","sequence":"additional","affiliation":[{"name":"IEIT SYSTEMS Company, Ltd., Jinan, China"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"ORCID":"https:\/\/orcid.org\/0009-0008-4460-9869","authenticated-orcid":false,"given":"Rengang","family":"Li","sequence":"additional","affiliation":[{"name":"IEIT SYSTEMS Company, Ltd., Jinan, China"}],"role":[{"vocabulary":"crossref","role":"author"}]}],"member":"263","reference":[{"key":"ref1","article-title":"OPT: Open pre-trained transformer language models","author":"Zhang","year":"2022"},{"key":"ref2","article-title":"GLM-130b: An open bilingual pre-trained model","author":"Zeng","year":"2022"},{"key":"ref4","article-title":"LLaMA: Open and efficient foundation language models","author":"Touvron","year":"2023"},{"key":"ref5","doi-asserted-by":"publisher","DOI":"10.1109\/ICCV51070.2023.00371"},{"key":"ref6","first-page":"1877","article-title":"Language models are few-shot learners","volume-title":"Proc. Adv. Neural Inf. Process. Syst.","author":"Brown"},{"key":"ref7","article-title":"A survey of large language models","author":"Zhao","year":"2023"},{"key":"ref8","article-title":"TensorFlow: Large-scale machine learning on heterogeneous distributed systems","author":"Abadi","year":"2016"},{"key":"ref9","first-page":"10435","article-title":"Mesh-tensorFlow: Deep learning for supercomputers","volume-title":"Proc. Adv. Neural Inf. Process. Syst.","author":"Shazeer"},{"key":"ref10","first-page":"8026","article-title":"PyTorch: An imperative style, high-performance deep learning library","volume-title":"Proc. Adv. Neural Inf. Process. Syst.","author":"Paszke"},{"key":"ref12","article-title":"Scaling language models: Methods, analysis & insights from training gopher","author":"Rae","year":"2021"},{"key":"ref14","article-title":"Training a 1 trillion parameter model with PyTorch fully sharded data parallel on AWS","year":"2022"},{"key":"ref15","doi-asserted-by":"publisher","DOI":"10.1145\/3341301.3359642"},{"key":"ref16","doi-asserted-by":"publisher","DOI":"10.1145\/3377454"},{"key":"ref17","doi-asserted-by":"crossref","first-page":"1","DOI":"10.1145\/3363554","article-title":"Scalable deep learning on distributed infrastructures: Challenges, techniques, and tools","volume":"53","author":"Mayer","year":"2020","journal-title":"ACM Comput. Surv."},{"key":"ref18","article-title":"Communication-efficient distributed deep learning: A comprehensive survey","author":"Tang","year":"2020"},{"key":"ref19","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":"ref20","article-title":"Horovod: Fast and easy distributed deep learning in tensorflow","author":"Sergeev","year":"2018"},{"key":"ref21","doi-asserted-by":"publisher","DOI":"10.1109\/IPDPS.2007.370405"},{"key":"ref22","doi-asserted-by":"publisher","DOI":"10.1006\/jpdc.1994.1085"},{"key":"ref23","first-page":"1223","article-title":"More effective distributed ML via a stale synchronous parallel parameter server","volume-title":"Proc. Adv. Neural Inf. Process. Syst.","author":"Ho"},{"key":"ref24","first-page":"693","article-title":"HOGWILD!: A lock-free approach to parallelizing stochastic gradient descent","volume-title":"Proc. Adv. Neural Inf. Process. Syst.","author":"Recht"},{"key":"ref25","first-page":"463","article-title":"A unified architecture for accelerating distributed DNN training in heterogeneous GPU\/CPU clusters","volume-title":"Proc. 14th USENIX Conf. Operating Syst. Des. Implementation","author":"Jiang"},{"key":"ref26","doi-asserted-by":"publisher","DOI":"10.48550\/arxiv.1811.06965"},{"key":"ref27","doi-asserted-by":"publisher","DOI":"10.1145\/3437801.3441593"},{"key":"ref28","doi-asserted-by":"publisher","DOI":"10.1145\/3458817.3476145"},{"key":"ref29","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":"ref30","article-title":"Towards understanding mixture of experts in deep learning","author":"Chen","year":"2022"},{"key":"ref31","doi-asserted-by":"publisher","DOI":"10.1145\/3394486.3406703"},{"key":"ref32","article-title":"Megatron-LM: Training multi-billion parameter language models using model parallelism","author":"Shoeybi","year":"2019"},{"key":"ref33","doi-asserted-by":"publisher","DOI":"10.1109\/TPDS.2023.3281931"},{"key":"ref34","doi-asserted-by":"publisher","DOI":"10.1145\/3302424.3303953"},{"key":"ref35","first-page":"559","article-title":"Alpa: Automating inter-and Intra-Operator parallelism for distributed deep learning","volume-title":"Proc. 16th USENIX Symp. Operating Syst. Des. Implementation","author":"Zheng"},{"key":"ref36","doi-asserted-by":"publisher","DOI":"10.1145\/3605573.3605613"},{"key":"ref37","doi-asserted-by":"publisher","DOI":"10.1109\/TPDS.2021.3132413"},{"key":"ref38","first-page":"267","article-title":"Unity: Accelerating DNN training through joint optimization of algebraic transformations and parallelization","volume-title":"Proc. 16th USENIX Symp. Operating Syst. Des. Implementation","author":"Unger"},{"key":"ref39","article-title":"Mixed precision training","author":"Micikevicius","year":"2017"},{"key":"ref40","doi-asserted-by":"publisher","DOI":"10.1109\/TETCI.2022.3220224"},{"key":"ref41","doi-asserted-by":"publisher","DOI":"10.1145\/3341301.3359630"},{"key":"ref42","first-page":"578","article-title":"TVM: An automated end-to-end optimizing compiler for deep learning","volume-title":"Proc. 13th USENIX Conf. Operating Syst. Des. Implementation","author":"Chen"},{"key":"ref43","doi-asserted-by":"publisher","DOI":"10.1109\/CCGRID.2019.00057"},{"key":"ref44","article-title":"TACOS: Topology-aware collective algorithm synthesizer for distributed training","author":"Won","year":"2023"},{"key":"ref45","first-page":"652","article-title":"On the utility of gradient compression in distributed training systems","volume-title":"Proc. Mach. Learn. Syst.","volume":"4","author":"Agarwal"},{"key":"ref46","first-page":"1027","article-title":"Accelerating collective communication in data parallel training across deep learning frameworks","volume-title":"Proc. 19th USENIX Symp. Netw. Syst. Des. Implementation","author":"Romero"},{"key":"ref47","doi-asserted-by":"publisher","DOI":"10.1109\/SC41405.2020.00024"},{"key":"ref48","first-page":"551","article-title":"ZeRO-offload: Democratizing billion-scale model training","volume-title":"Proc. USENIX Annu. Tech. Conf.","author":"Ren"},{"key":"ref49","doi-asserted-by":"publisher","DOI":"10.1145\/3458817.3476205"},{"key":"ref50","first-page":"497","article-title":"Checkmate: Breaking the memory wall with optimal tensor rematerialization","volume-title":"Proc. Mach. Learn. Syst.","volume":"2","author":"Jain"},{"key":"ref51","article-title":"Training deep nets with sublinear memory cost","author":"Chen","year":"2016"},{"key":"ref52","first-page":"341","article-title":"Reducing activation recomputation in large transformer models","volume-title":"Proc. Mach. Learn. Syst.","author":"Korthikanti"},{"key":"ref53","doi-asserted-by":"publisher","DOI":"10.1109\/ISCA.2018.00070"},{"key":"ref54","doi-asserted-by":"publisher","DOI":"10.1109\/HCS55958.2022.9895599"},{"key":"ref55","first-page":"287","article-title":"Direct access,high-performance memory disaggregation with DirectCXL","volume-title":"Proc. USENIX Annu. Tech. Conf.","author":"Gouk"},{"key":"ref56","doi-asserted-by":"publisher","DOI":"10.1145\/2258996.2259012"},{"key":"ref57","doi-asserted-by":"publisher","DOI":"10.1145\/3545008.3545090"},{"key":"ref58","doi-asserted-by":"publisher","DOI":"10.1109\/MSST.2010.5496972"},{"key":"ref59","doi-asserted-by":"publisher","DOI":"10.14778\/3446095.3446100"},{"key":"ref60","first-page":"33","article-title":"Plumber: Diagnosing and removing performance bottlenecks in machine learning data pipelines","volume-title":"Proc. Mach. Learn. Syst.","author":"Kuchnik"},{"key":"ref61","doi-asserted-by":"publisher","DOI":"10.1016\/j.future.2018.04.073"},{"key":"ref62","article-title":"Application-specific hardware accelerators","author":"Arnold","year":"2019"},{"key":"ref63","doi-asserted-by":"publisher","DOI":"10.1145\/3079856.3080246"},{"key":"ref64","doi-asserted-by":"publisher","DOI":"10.1587\/transinf.2020EDP7201"},{"key":"ref65","doi-asserted-by":"publisher","DOI":"10.1109\/ISPASS57527.2023.00035"},{"key":"ref66","doi-asserted-by":"publisher","DOI":"10.1109\/ISPASS48437.2020.00018"},{"key":"ref67","doi-asserted-by":"publisher","DOI":"10.1109\/ISCA52012.2021.00049"},{"key":"ref68","doi-asserted-by":"publisher","DOI":"10.1109\/TPDS.2021.3062721"},{"key":"ref69","doi-asserted-by":"publisher","DOI":"10.1155\/2023\/2663115"},{"key":"ref70","article-title":"The history began from alexnet: A comprehensive survey on deep learning approaches","author":"Alom","year":"2018"},{"key":"ref71","article-title":"Very deep convolutional networks for large-scale image recognition","author":"Simonyan","year":"2014"},{"key":"ref72","doi-asserted-by":"publisher","DOI":"10.1109\/ISPASS57527.2023.00037"},{"key":"ref73","doi-asserted-by":"publisher","DOI":"10.1145\/3190508.3190517"},{"key":"ref74","doi-asserted-by":"publisher","DOI":"10.1109\/DASC\/PiCom\/DataCom\/CyberSciTec.2018.000-4"},{"key":"ref75","doi-asserted-by":"publisher","DOI":"10.1109\/IUCC-CIT-DSCI-SmartCNS55181.2021.00071"},{"key":"ref76","doi-asserted-by":"publisher","DOI":"10.1109\/TPDS.2022.3151739"},{"key":"ref77","doi-asserted-by":"publisher","DOI":"10.1007\/s10766-022-00741-6"},{"key":"ref78","doi-asserted-by":"publisher","DOI":"10.1145\/3503221.3508418"},{"key":"ref79","article-title":"BERT: Pre-training of deep bidirectional transformers for language understanding","author":"Devlin","year":"2018"},{"key":"ref80","doi-asserted-by":"publisher","DOI":"10.1145\/3581784.3607102"},{"key":"ref81","doi-asserted-by":"publisher","DOI":"10.1145\/2783258.2783270"},{"key":"ref82","first-page":"412","article-title":"Distributed hierarchical GPU parameter server for massive scale deep learning ADS systems","volume-title":"Proc. Mach. Learn. Syst.","author":"Zhao"},{"key":"ref83","doi-asserted-by":"publisher","DOI":"10.1016\/j.jpdc.2008.09.002"},{"key":"ref84","doi-asserted-by":"publisher","DOI":"10.48550\/ARXIV.1706.03762"},{"key":"ref85","doi-asserted-by":"publisher","DOI":"10.1145\/3337821.3337873"},{"key":"ref86","doi-asserted-by":"publisher","DOI":"10.1109\/MASCOTS56607.2022.00011"},{"key":"ref87","first-page":"2274","article-title":"Exploring hidden dimensions in accelerating convolutional neural networks","volume-title":"Proc. Int. Conf. Mach. Learn.","author":"Jia"},{"key":"ref88","doi-asserted-by":"publisher","DOI":"10.1109\/ACCESS.2019.2916550"},{"key":"ref89","doi-asserted-by":"publisher","DOI":"10.1109\/BigData.2016.7840590"},{"key":"ref90","article-title":"Paleo: A performance model for deep neural networks","volume-title":"Proc. Int. Conf. Learn. Representations","author":"Qi"},{"key":"ref91","article-title":"Network in network","author":"Lin","year":"2013"},{"key":"ref92","doi-asserted-by":"publisher","DOI":"10.5555\/3298023.3298188"},{"key":"ref93","first-page":"1","article-title":"Beyond data and model parallelism for deep neural networks","volume-title":"Proc. Mach. Learn. Syst.","author":"Jia"},{"key":"ref94","doi-asserted-by":"publisher","DOI":"10.18653\/v1\/D15-1166"},{"key":"ref95","first-page":"337","article-title":"Daydream: Accurately estimating the efficacy of optimizations for DNN training","volume-title":"Proc. USENIX Annu. Tech. Conf.","author":"Zhu"},{"key":"ref96","doi-asserted-by":"publisher","DOI":"10.1109\/CVPR.2016.90"},{"key":"ref97","article-title":"Google\u2019s neural machine translation system: Bridging the gap between human and machine translation","author":"Wu","year":"2016"},{"key":"ref98","first-page":"623","article-title":"dPRO: A generic performance diagnosis and optimization toolkit for expediting distributed DNN training","volume-title":"Proc. Mach. Learn. Syst.","author":"Hu"},{"key":"ref99","doi-asserted-by":"publisher","DOI":"10.1145\/3587135.3592200"},{"key":"ref100","article-title":"Language models are unsupervised multitask learners","volume":"1","author":"Radford","year":"2019","journal-title":"OpenAI blog"},{"key":"ref101","first-page":"1","article-title":"Exploring the limits of transfer learning with a unified text-to-text transformer","volume":"21","author":"Raffel","year":"2020","journal-title":"J. Mach. Learn. Res."},{"key":"ref102","doi-asserted-by":"publisher","DOI":"10.1109\/tpds.2024.3443255"},{"key":"ref103","article-title":"Deep learning recommendation model for personalization and recommendation systems","author":"Naumov","year":"2019"},{"key":"ref104","doi-asserted-by":"publisher","DOI":"10.1109\/TPDS.2023.3243261"},{"key":"ref105","doi-asserted-by":"publisher","DOI":"10.1080\/00031305.1995.10476177"},{"key":"ref107","first-page":"132","article-title":"Priority-based parameter propagation for distributed DNN training","volume-title":"Proc. Mach. Learn. Syst.","volume":"1","author":"Jayarajan"},{"key":"ref108","article-title":"NVIDIA, API documentation of NVIDIA\u2019s apex optimizers","year":"2018"},{"key":"ref109","doi-asserted-by":"publisher","DOI":"10.1145\/368996.369025"},{"key":"ref110","doi-asserted-by":"publisher","DOI":"10.1016\/S0167-8191(06)80021-9"},{"key":"ref111","doi-asserted-by":"publisher","DOI":"10.1016\/j.aiopen.2021.01.001"},{"key":"ref112","doi-asserted-by":"publisher","DOI":"10.1007\/11871842_29"},{"key":"ref113","article-title":"Distributed machine learning via sufficient factor broadcasting","author":"Xie","year":"2015"},{"key":"ref114","doi-asserted-by":"publisher","DOI":"10.1109\/MASCOTS.2018.00037"},{"key":"ref115","doi-asserted-by":"publisher","DOI":"10.1109\/TCOM.1979.1094531"},{"key":"ref116","doi-asserted-by":"publisher","DOI":"10.1145\/3358960.3379141"},{"key":"ref117","doi-asserted-by":"publisher","DOI":"10.1109\/CloudIntelligence52565.2021.00012"},{"key":"ref118","doi-asserted-by":"publisher","DOI":"10.1016\/j.jpdc.2014.06.008"},{"key":"ref119","first-page":"722","article-title":"DNNEmu: A lightweight performance emulator for distributed DNN training","volume-title":"Proc. Springer Int. Conf. Algorithms Archit. Parallel Process.","author":"Wang"},{"key":"ref120","first-page":"3149","article-title":"LightGBM: A highly efficient gradient boosting decision tree","volume-title":"Proc. Adv. Neural Inf. Process. Syst.","author":"Ke"},{"key":"ref121","article-title":"MXNet: A. flexible and efficient machine learning library for heterogeneous distributed systems","author":"Chen","year":"2015"},{"key":"ref122","doi-asserted-by":"publisher","DOI":"10.1109\/BigData.2018.8622396"},{"key":"ref123","first-page":"503","article-title":"Habitat: A runtime-based computational performance predictor for deep neural network training","volume-title":"Proc. 2021 USENIX Annu. Techn. Conf.","author":"Geoffrey"},{"key":"ref124","doi-asserted-by":"publisher","DOI":"10.1109\/ICSE-SEIP58684.2023.00039"},{"key":"ref125","doi-asserted-by":"publisher","DOI":"10.1145\/3437984.3458829"},{"key":"ref126","article-title":"MLIR: A compiler infrastructure for the end of Moore\u2019s law","author":"Lattner","year":"2020"},{"key":"ref127","article-title":"SCALE-sim: Systolic CNN accelerator simulator","author":"Samajdar","year":"2018"},{"key":"ref128","doi-asserted-by":"publisher","DOI":"10.1109\/ISPASS.2009.4919636"},{"key":"ref129","article-title":"Network simulations with the NS-3 simulator","volume":"14","author":"Henderson","year":"2008","journal-title":"SIGCOMM Demonstration"},{"key":"ref130","doi-asserted-by":"publisher","DOI":"10.2307\/j.ctt20mvcpx.51"},{"key":"ref131","doi-asserted-by":"publisher","DOI":"10.1007\/978-3-030-85665-6_13"},{"key":"ref132","article-title":"COMET: A comprehensive cluster design methodology for distributed deep learning training","author":"Kadiyala","year":"2022"},{"key":"ref133","article-title":"Restructuring, pruning, and adjustment of deep models for parallel distributed inference","author":"Abdi","year":"2020"},{"key":"ref134","doi-asserted-by":"publisher","DOI":"10.1201\/9780429399602"},{"key":"ref135","doi-asserted-by":"publisher","DOI":"10.1109\/TWC.2021.3088910"},{"key":"ref136","first-page":"161","article-title":"Zico: Efficient GPU memory sharing for concurrent DNN training","volume-title":"Proc. USENIX Annu. Techn. Conf.","author":"Lim"},{"key":"ref137","doi-asserted-by":"publisher","DOI":"10.1109\/ISPASS.2009.4919648"},{"key":"ref138","article-title":"The AMD gem5 APU simulator: Modeling heterogeneous systems in gem5","volume-title":"Proc. Tut. Int. Symp. Microarchit.","author":"Beckmann"},{"key":"ref139","doi-asserted-by":"publisher","DOI":"10.1109\/LCA.2020.2973991"},{"key":"ref140","doi-asserted-by":"publisher","DOI":"10.1145\/324138.324232"},{"key":"ref141","doi-asserted-by":"publisher","DOI":"10.1109\/MM.2020.2974843"},{"key":"ref142","doi-asserted-by":"publisher","DOI":"10.1145\/3544216.3544248"},{"key":"ref143","doi-asserted-by":"publisher","DOI":"10.1109\/MNET.013.2300053"},{"key":"ref144","doi-asserted-by":"publisher","DOI":"10.1109\/ICS51289.2020.00039"},{"key":"ref145","doi-asserted-by":"publisher","DOI":"10.1109\/TPDS.2021.3052895"},{"key":"ref146","first-page":"1676","article-title":"SpotLight: Optimizing device placement for training deep neural networks","volume-title":"Proc. Int. Conf. Mach. Learn.","author":"Gao"}],"container-title":["IEEE Transactions on Parallel and Distributed Systems"],"original-title":[],"link":[{"URL":"http:\/\/xplorestaging.ieee.org\/ielx8\/71\/10719599\/10707191.pdf?arnumber=10707191","content-type":"unspecified","content-version":"vor","intended-application":"similarity-checking"}],"deposited":{"date-parts":[[2024,11,27]],"date-time":"2024-11-27T00:40:45Z","timestamp":1732668045000},"score":1,"resource":{"primary":{"URL":"https:\/\/ieeexplore.ieee.org\/document\/10707191\/"}},"subtitle":[],"short-title":[],"issued":{"date-parts":[[2024,12]]},"references-count":142,"journal-issue":{"issue":"12"},"URL":"https:\/\/doi.org\/10.1109\/tpds.2024.3476390","relation":{},"ISSN":["1045-9219","1558-2183","2161-9883"],"issn-type":[{"value":"1045-9219","type":"print"},{"value":"1558-2183","type":"electronic"},{"value":"2161-9883","type":"electronic"}],"subject":[],"published":{"date-parts":[[2024,12]]}}}