{"status":"ok","message-type":"work","message-version":"1.0.0","message":{"indexed":{"date-parts":[[2026,6,4]],"date-time":"2026-06-04T16:05:16Z","timestamp":1780589116980,"version":"3.54.1"},"reference-count":68,"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"}],"content-domain":{"domain":[],"crossmark-restriction":false},"short-container-title":["IEEE Trans. Netw."],"published-print":{"date-parts":[[2026]]},"DOI":"10.1109\/ton.2025.3645806","type":"journal-article","created":{"date-parts":[[2025,12,18]],"date-time":"2025-12-18T18:32:21Z","timestamp":1766082741000},"page":"2478-2497","source":"Crossref","is-referenced-by-count":3,"title":["Optimizing Mixture-of-Experts Inference Time via Model Deployment and Communication Scheduling"],"prefix":"10.1109","volume":"34","author":[{"ORCID":"https:\/\/orcid.org\/0000-0003-3416-5551","authenticated-orcid":false,"given":"Jialong","family":"Li","sequence":"first","affiliation":[{"name":"Shenzhen University of Advanced Technology, Shenzhen, China"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"ORCID":"https:\/\/orcid.org\/0000-0001-9518-7557","authenticated-orcid":false,"given":"Shreyansh","family":"Tripathi","sequence":"additional","affiliation":[{"name":"Max Planck Institute for Informatics, Saarbrücken, Germany"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Lakshay","family":"Rastogi","sequence":"additional","affiliation":[{"name":"Max Planck Institute for Informatics, Saarbrücken, Germany"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Yiming","family":"Lei","sequence":"additional","affiliation":[{"name":"Max Planck Institute for Informatics, Saarbrücken, Germany"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"ORCID":"https:\/\/orcid.org\/0000-0002-6973-3259","authenticated-orcid":false,"given":"Rui","family":"Pan","sequence":"additional","affiliation":[{"name":"Princeton University, Princeton, USA"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"ORCID":"https:\/\/orcid.org\/0000-0003-2981-1459","authenticated-orcid":false,"given":"Yiting","family":"Xia","sequence":"additional","affiliation":[{"name":"Max Planck Institute for Informatics, Saarbrücken, Germany"}],"role":[{"vocabulary":"crossref","role":"author"}]}],"member":"263","reference":[{"key":"ref1","doi-asserted-by":"publisher","DOI":"10.1109\/SC41405.2020.00024"},{"key":"ref2","doi-asserted-by":"publisher","DOI":"10.14778\/3503585.3503590"},{"key":"ref3","first-page":"463","article-title":"A unified architecture for accelerating distributed DNN training in heterogeneous GPU\/CPU clusters","volume-title":"Proc. 14th USENIX Symp. Operating Syst. Design Implement. (OSDI)","author":"Jiang"},{"key":"ref4","doi-asserted-by":"publisher","DOI":"10.1145\/3563766.3564096"},{"key":"ref5","doi-asserted-by":"publisher","DOI":"10.14778\/3415478.3415530"},{"key":"ref6","article-title":"Megatron-LM: Training multi-billion parameter language models using model parallelism","author":"Shoeybi","year":"2019","journal-title":"arXiv:1909.08053"},{"key":"ref7","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":"ref8","article-title":"Outrageously large neural networks: The sparsely-gated mixture-of-experts layer","author":"Shazeer","year":"2017","journal-title":"arXiv:1701.06538"},{"key":"ref9","doi-asserted-by":"publisher","DOI":"10.1145\/3503221.3508418"},{"key":"ref10","article-title":"Towards MoE deployment: Mitigating inefficiencies in mixture-of-expert (MoE) inference","author":"Huang","year":"2023","journal-title":"arXiv:2303.06182"},{"key":"ref11","first-page":"945","article-title":"Accelerating distributed MoE training and inference with lina","volume-title":"Proc. USENIX Annu. Tech. Conf. (USENIX ATC)","author":"Li"},{"key":"ref12","doi-asserted-by":"publisher","DOI":"10.1109\/tsc.2024.3399654"},{"key":"ref13","doi-asserted-by":"publisher","DOI":"10.1145\/3588964"},{"key":"ref14","doi-asserted-by":"publisher","DOI":"10.1145\/3603269.3604869"},{"key":"ref15","doi-asserted-by":"publisher","DOI":"10.1109\/TSC.2024.3399654"},{"key":"ref16","article-title":"Lazarus: Resilient and elastic training of mixture-of-experts models","author":"Wu","year":"2024","journal-title":"arXiv:2407.04656"},{"key":"ref17","doi-asserted-by":"publisher","DOI":"10.1145\/3627703.3650083"},{"key":"ref18","first-page":"945","article-title":"MLaaS in the wild: Workload analysis and scheduling in large-scale heterogeneous GPU clusters","volume-title":"Proc. 19th USENIX Symp. Networked Syst. Design Implement. (NSDI)","author":"Weng"},{"key":"ref19","first-page":"22173","article-title":"TA-MoE: Topology-aware large scale mixture-of-expert training","volume-title":"Proc. Adv. Neural Inf. Process. Syst.","author":"Chen"},{"key":"ref20","doi-asserted-by":"publisher","DOI":"10.1145\/3386367.3432728"},{"key":"ref21","first-page":"481","article-title":"Heterogeneity-aware cluster scheduling policies for deep learning workloads","volume-title":"Proc. 14th USENIX Symp. Operating Syst. Design Implement. (OSDI)","author":"Narayanan"},{"issue":"120","key":"ref22","first-page":"1","article-title":"Switch transformers: Scaling to trillion parameter models with simple and efficient sparsity","volume":"23","author":"Fedus","year":"2021","journal-title":"J. Mach. Learn. Res."},{"key":"ref23","article-title":"GShard: Scaling giant models with conditional computation and automatic sharding","author":"Lepikhin","year":"2020","journal-title":"arXiv:2006.16668"},{"key":"ref24","first-page":"8583","article-title":"Scaling vision with sparse mixture of experts","volume-title":"Proc. NIPS","author":"Riquelme"},{"key":"ref25","article-title":"Generating long sequences with sparse transformers","author":"Child","year":"2019","journal-title":"arXiv:1904.10509"},{"key":"ref26","first-page":"269","article-title":"Tutel: Adaptive mixture-of-experts at scale","volume":"5","author":"Hwang","year":"2022","journal-title":"Proc. Mach. Learn. Syst."},{"key":"ref27","article-title":"Pre-gated MoE: An algorithm-system co-design for fast and scalable mixture-of-expert inference","author":"Hwang","year":"2023","journal-title":"arXiv:2308.12066"},{"key":"ref28","article-title":"HetuMoE: An efficient trillion-scale mixture-of-expert distributed training system","author":"Nie","year":"2022","journal-title":"arXiv:2203.14685"},{"key":"ref29","article-title":"FastMoE: A fast mixture-of-expert training system","author":"He","year":"2021","journal-title":"arXiv:2103.13262"},{"key":"ref30","doi-asserted-by":"publisher","DOI":"10.1109\/CLUSTER52292.2023.00015"},{"key":"ref31","article-title":"Deepep: An efficient expert-parallel communication library","author":"Zhao","year":"2025"},{"key":"ref32","article-title":"Improving language understanding by generative pre-training","author":"Radford","year":"2018","journal-title":"arXiv:2012.11747v3"},{"key":"ref33","article-title":"Qwen3 technical report","volume-title":"arXiv:2505.09388","author":"Yang","year":"2025"},{"key":"ref34","article-title":"DeepSeek-V3 technical report","volume-title":"arXiv:2412.19437","author":"Liu","year":"2024"},{"key":"ref35","article-title":"Mixtral of experts","author":"Jiang","year":"2024","journal-title":"arXiv:2401.04088"},{"key":"ref36","first-page":"533","article-title":"${\\$}AntMan{\\$}$\n: Dynamic scaling on GPU clusters for deep learning","volume-title":"Proc. 14th USENIX Symp. Operating Syst. Design Implement. (OSDI 20)","author":"Xiao"},{"key":"ref37","article-title":"Boosting LLM serving through spatial\u2013temporal GPU resource sharing","author":"Lin","year":"2025","journal-title":"arXiv:2504.19516"},{"key":"ref38","first-page":"193","article-title":"DistServe: Disaggregating prefill and decoding for goodput-optimized large language model serving","volume-title":"Proc. 18th USENIX Symp. Operating Syst. Design Implement. (OSDI)","author":"Zhong"},{"key":"ref39","article-title":"SARATHI: Efficient LLM inference by piggybacking decodes with chunked prefills","author":"Agrawal","year":"2023","journal-title":"arXiv:2308.16369"},{"key":"ref40","doi-asserted-by":"publisher","DOI":"10.1145\/3600006.3613175"},{"key":"ref41","doi-asserted-by":"publisher","DOI":"10.1145\/3627703.3629580"},{"key":"ref42","first-page":"161","article-title":"Zico: Efficient GPU memory sharing for concurrent DNN training","volume-title":"Proc. USENIX Annu. Tech. Conf. (USENIX ATC)","author":"Lim"},{"key":"ref43","doi-asserted-by":"publisher","DOI":"10.1145\/2716281.2836086"},{"key":"ref44","doi-asserted-by":"publisher","DOI":"10.1145\/3544216.3544235"},{"key":"ref45","doi-asserted-by":"publisher","DOI":"10.1145\/2619239.2626309"},{"key":"ref46","doi-asserted-by":"publisher","DOI":"10.1145\/3098822.3098825"},{"key":"ref47","doi-asserted-by":"publisher","DOI":"10.1145\/3230543.3230564"},{"key":"ref48","article-title":"MoETuner: Optimized mixture of expert serving with balanced expert placement and token routing","author":"Go","year":"2025","journal-title":"arXiv:2502.06643"},{"key":"ref49","volume-title":"Farkas\u2019 Lemma","year":"2024"},{"key":"ref50","doi-asserted-by":"publisher","DOI":"10.1145\/3190508.3190517"},{"key":"ref51","first-page":"485","article-title":"Tiresias: A GPU cluster manager for distributed deep learning","volume-title":"Proc. 16th USENIX Symp. Networked Syst. Design Implement. (NSDI)","author":"Gu"},{"key":"ref52","first-page":"289","article-title":"Themis: Fair and efficient GPU cluster scheduling","volume-title":"Proc. 17th USENIX Symp. Networked Syst. Design Implement. (NSDI)","author":"Mahajan"},{"key":"ref53","first-page":"1","article-title":"Pollux: Co-adaptive cluster scheduling for goodput-optimized deep learning","volume-title":"Proc. 15th USENIX Symp. Operating Syst. Design Implement. (OSDI)","author":"Qiao"},{"key":"ref54","first-page":"721","article-title":"Elastic resource sharing for distributed deep learning","volume-title":"Proc. 18th USENIX Symp. Networked Syst. Design Implement. (NSDI)","author":"Hwang"},{"key":"ref55","article-title":"Synergy: Resource sensitive DNN scheduling in multi-tenant clusters","author":"Mohan","year":"2021","journal-title":"arXiv:2110.06073"},{"key":"ref56","doi-asserted-by":"publisher","DOI":"10.1007\/978-3-642-51576-7_5"},{"key":"ref57","volume-title":"The Art of Computer Programming: Sorting and Searching","volume":"3","author":"Knuth","year":"1998"},{"key":"ref58","doi-asserted-by":"publisher","DOI":"10.1137\/0202019"},{"key":"ref59","doi-asserted-by":"publisher","DOI":"10.1016\/0377-2217(92)90078-N"},{"key":"ref60","first-page":"9564","article-title":"Multimodal contrastive learning with LIMoE: The language-image mixture of experts","volume-title":"Proc. Adv. Neural Inf. Process. Syst.","author":"Mustafa"},{"key":"ref61","article-title":"LLM-jp: A cross-organizational project for the research and development of fully open Japanese LLMs","author":"Aizawa","year":"2024","journal-title":"arXiv:2407.03963"},{"key":"ref62","article-title":"The stack: 3 TB of permissively licensed source code","author":"Kocetkov","year":"2022","journal-title":"arXiv:2211.15533"},{"key":"ref63","article-title":"SYMI: Efficient mixture-of-experts training via model and optimizer state decoupling","author":"Skiadopoulos","year":"2025","journal-title":"arXiv:2504.19925"},{"key":"ref64","first-page":"595","article-title":"Gandiva: Introspective cluster scheduling for deep learning","volume-title":"Proc. 13th USENIX Symp. Oper. Syst. Design Implement. (OSDI)","author":"Xiao"},{"key":"ref65","first-page":"98","article-title":"Salus: Fine-grained GPU sharing primitives for deep learning applications","volume-title":"Proc. MLSys","author":"Yu"},{"key":"ref66","first-page":"696","article-title":"Wavelet: Efficient DNN training with tick-tock scheduling","volume-title":"Proc. Mach. Learn. Syst.","volume":"3","author":"Wang"},{"key":"ref67","doi-asserted-by":"publisher","DOI":"10.1145\/181014.181756"},{"key":"ref68","doi-asserted-by":"publisher","DOI":"10.1145\/3502181.3531468"}],"container-title":["IEEE Transactions on Networking"],"original-title":[],"link":[{"URL":"http:\/\/xplorestaging.ieee.org\/ielx8\/10723154\/11317935\/11303974.pdf?arnumber=11303974","content-type":"unspecified","content-version":"vor","intended-application":"similarity-checking"}],"deposited":{"date-parts":[[2026,1,13]],"date-time":"2026-01-13T08:11:04Z","timestamp":1768291864000},"score":1,"resource":{"primary":{"URL":"https:\/\/ieeexplore.ieee.org\/document\/11303974\/"}},"subtitle":[],"short-title":[],"issued":{"date-parts":[[2026]]},"references-count":68,"URL":"https:\/\/doi.org\/10.1109\/ton.2025.3645806","relation":{},"ISSN":["2998-4157"],"issn-type":[{"value":"2998-4157","type":"electronic"}],"subject":[],"published":{"date-parts":[[2026]]}}}