{"status":"ok","message-type":"work","message-version":"1.0.0","message":{"indexed":{"date-parts":[[2026,6,4]],"date-time":"2026-06-04T16:04:17Z","timestamp":1780589057295,"version":"3.54.1"},"reference-count":69,"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"}],"funder":[{"name":"Guangdong Provincial Young Innovative Talent Program"}],"content-domain":{"domain":[],"crossmark-restriction":false},"short-container-title":["IEEE Trans. Netw."],"published-print":{"date-parts":[[2026]]},"DOI":"10.1109\/ton.2026.3676382","type":"journal-article","created":{"date-parts":[[2026,3,23]],"date-time":"2026-03-23T20:06:05Z","timestamp":1774296365000},"page":"4431-4448","source":"Crossref","is-referenced-by-count":1,"title":["RailS: Load Balancing for All-to-All Communication in Distributed Mixture-of-Experts Training"],"prefix":"10.1109","volume":"34","author":[{"ORCID":"https:\/\/orcid.org\/0009-0001-1599-8200","authenticated-orcid":false,"given":"Heng","family":"Xu","sequence":"first","affiliation":[{"name":"Faculty of Computer Science and Artificial Intelligence, Shenzhen University of Advanced Technology, Shenzhen, China"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"ORCID":"https:\/\/orcid.org\/0009-0009-7776-4985","authenticated-orcid":false,"given":"Zhiwei","family":"Yu","sequence":"additional","affiliation":[{"name":"Institute for Network Sciences and Cyberspace, Tsinghua University, Beijing, China"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"ORCID":"https:\/\/orcid.org\/0009-0005-5313-7750","authenticated-orcid":false,"given":"Chengze","family":"Du","sequence":"additional","affiliation":[{"name":"Faculty of Computer Science and Artificial Intelligence, Shenzhen University of Advanced Technology, Shenzhen, China"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"ORCID":"https:\/\/orcid.org\/0009-0001-9803-262X","authenticated-orcid":false,"given":"Ying","family":"Zhou","sequence":"additional","affiliation":[{"name":"Faculty of Computer Science and Artificial Intelligence, Shenzhen University of Advanced Technology, Shenzhen, China"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"ORCID":"https:\/\/orcid.org\/0000-0001-6856-1785","authenticated-orcid":false,"given":"Letian","family":"Li","sequence":"additional","affiliation":[{"name":"Department of Information Engineering, The Chinese University of Hong Kong, Shatin, Hong Kong"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"ORCID":"https:\/\/orcid.org\/0009-0001-2721-7475","authenticated-orcid":false,"given":"Haojie","family":"Wang","sequence":"additional","affiliation":[{"name":"China Mobile, Beijing, China"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Weiqiang","family":"Cheng","sequence":"additional","affiliation":[{"name":"China Mobile, Beijing, China"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"ORCID":"https:\/\/orcid.org\/0000-0003-3416-5551","authenticated-orcid":false,"given":"Jialong","family":"Li","sequence":"additional","affiliation":[{"name":"Faculty of Computer Science and Artificial Intelligence, Shenzhen University of Advanced Technology, Shenzhen, China"}],"role":[{"vocabulary":"crossref","role":"author"}]}],"member":"263","reference":[{"key":"ref1","volume-title":"Infiniband Architecture Specification Release 1.2.1 Annex A16: Roce","year":"2010"},{"key":"ref2","volume-title":"Infiniband Architecture Specification Release 1.2.1 Annex A17: Rocev2","year":"2014"},{"key":"ref3","doi-asserted-by":"publisher","DOI":"10.1145\/1402946.1402967"},{"key":"ref4","first-page":"89","article-title":"Hedera: Dynamic flow scheduling for data center networks","volume-title":"Proc. 7th USENIX Conf. Networked Syst. Design Implement.","author":"Al-Fares"},{"key":"ref5","doi-asserted-by":"publisher","DOI":"10.1145\/2619239.2626316"},{"key":"ref6","doi-asserted-by":"publisher","DOI":"10.1145\/2534169.2486031"},{"key":"ref7","doi-asserted-by":"publisher","DOI":"10.1145\/2079296.2079304"},{"key":"ref8","article-title":"REPS: Recycled entropy packet spraying for adaptive load balancing and failure mitigation","author":"Bonato","year":"2024","journal-title":"arXiv:2407.21625"},{"key":"ref9","first-page":"1877","article-title":"Language models are few-shot learners","volume-title":"Proc. NIPS","author":"Brown"},{"key":"ref10","doi-asserted-by":"publisher","DOI":"10.1145\/3544216.3544235"},{"key":"ref11","doi-asserted-by":"publisher","DOI":"10.48550\/arXiv.1810.04805"},{"key":"ref12","doi-asserted-by":"publisher","DOI":"10.1109\/INFCOM.2013.6567015"},{"key":"ref13","first-page":"5547","article-title":"GLaM: Efficient scaling of language models with mixture-of-experts","volume-title":"Proc. Int. Conf. Mach. Learn.","author":"Du"},{"issue":"120","key":"ref14","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":"ref15","doi-asserted-by":"publisher","DOI":"10.1145\/3651890.3672233"},{"key":"ref16","doi-asserted-by":"publisher","DOI":"10.1137\/0117039"},{"key":"ref17","doi-asserted-by":"publisher","DOI":"10.1145\/1592568.1592576"},{"key":"ref18","doi-asserted-by":"publisher","DOI":"10.1145\/1592568.1592577"},{"key":"ref19","doi-asserted-by":"publisher","DOI":"10.1145\/1402958.1402968"},{"key":"ref20","doi-asserted-by":"publisher","DOI":"10.1038\/s41586-025-09422-z"},{"key":"ref21","doi-asserted-by":"publisher","DOI":"10.1145\/3098822.3098825"},{"key":"ref22","doi-asserted-by":"publisher","DOI":"10.1109\/CVPR.2016.90"},{"key":"ref23","doi-asserted-by":"publisher","DOI":"10.1145\/2829988.2787507"},{"key":"ref24","doi-asserted-by":"publisher","DOI":"10.17487\/rfc2992"},{"key":"ref25","doi-asserted-by":"publisher","DOI":"10.1109\/TNET.2024.3366336"},{"key":"ref26","doi-asserted-by":"publisher","DOI":"10.1145\/3005745.3005760"},{"key":"ref27","article-title":"Mixtral of experts","author":"Jiang","year":"2024","journal-title":"arXiv:2401.04088"},{"key":"ref28","doi-asserted-by":"publisher","DOI":"10.1145\/2674005.2674985"},{"key":"ref29","doi-asserted-by":"publisher","DOI":"10.1145\/1232919.1232925"},{"key":"ref30","doi-asserted-by":"publisher","DOI":"10.1145\/3143361.3143401"},{"key":"ref31","doi-asserted-by":"publisher","DOI":"10.1145\/2890955.2890968"},{"key":"ref32","article-title":"STrack: A reliable multipath transport for AI\/ML clusters","author":"Le","year":"2024","journal-title":"arXiv:2407.15266"},{"key":"ref33","article-title":"GShard: Scaling giant models with conditional computation and automatic sharding","author":"Lepikhin","year":"2020","journal-title":"arXiv:2006.16668"},{"key":"ref34","doi-asserted-by":"publisher","DOI":"10.1109\/TON.2025.3590083"},{"key":"ref35","doi-asserted-by":"publisher","DOI":"10.1145\/3651890.3672245"},{"key":"ref36","doi-asserted-by":"publisher","DOI":"10.1145\/3542637.3542647"},{"key":"ref37","article-title":"Optimizing mixture-of-experts inference time combining model deployment and communication scheduling","author":"Li","year":"2024","journal-title":"arXiv:2410.17043"},{"key":"ref38","doi-asserted-by":"publisher","DOI":"10.1145\/3718958.3750465"},{"key":"ref39","article-title":"DeepSeek-V3 technical report","volume-title":"arXiv:2412.19437","author":"Liu","year":"2024"},{"key":"ref40","first-page":"357","article-title":"Multi-path transport for RDMA in datacenters","volume-title":"Proc. 15th USENIX Symp. Networked Syst. Design Implement.","author":"Lu"},{"key":"ref41","doi-asserted-by":"publisher","DOI":"10.1145\/3230543.3230557"},{"key":"ref42","doi-asserted-by":"publisher","DOI":"10.1109\/71.963420"},{"key":"ref43","volume-title":"SuperPOD: Next Generation Scalable Infrastructure for AI Leadership","year":"2023"},{"key":"ref44","first-page":"761","article-title":"An edge-queued datagram service for all datacenter traffic","volume-title":"Proc. 19th USENIX Symp. Networked Syst. Design Implement. (NSDI 22)","author":"Olteanu"},{"key":"ref45","doi-asserted-by":"publisher","DOI":"10.1145\/2630088.2631977"},{"key":"ref46","doi-asserted-by":"publisher","DOI":"10.1145\/2619239.2626309"},{"key":"ref47","doi-asserted-by":"publisher","DOI":"10.1145\/3651890.3672265"},{"key":"ref48","doi-asserted-by":"publisher","DOI":"10.1145\/3544216.3544226"},{"key":"ref49","doi-asserted-by":"publisher","DOI":"10.1145\/2043164.2018467"},{"key":"ref50","doi-asserted-by":"publisher","DOI":"10.1145\/1999916.1999919"},{"key":"ref51","first-page":"399","article-title":"How hard can it be? Designing and implementing a deployable multipath TCP","volume-title":"Proc. 9th USENIX Symp. Networked Syst. Design Implement. (NSDI)","author":"Raiciu"},{"key":"ref52","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":"ref53","doi-asserted-by":"publisher","DOI":"10.17487\/rfc5040"},{"key":"ref54","doi-asserted-by":"publisher","DOI":"10.1145\/2535372.2535397"},{"key":"ref55","doi-asserted-by":"publisher","DOI":"10.1145\/2785956.2787508"},{"key":"ref56","doi-asserted-by":"publisher","DOI":"10.1145\/3603269.3604849"},{"key":"ref57","article-title":"LLaMA: Open and efficient foundation language models","author":"Touvron","year":"2023","journal-title":"arXiv:2302.13971"},{"key":"ref58","first-page":"407","article-title":"Let it flow: Resilient asymmetric load balancing with flowlet switching","volume-title":"Proc. 14th USENIX Symp. Netw. Syst. Design Implement. (NSDI)","author":"Vanini"},{"key":"ref59","doi-asserted-by":"publisher","DOI":"10.1145\/3748273.3749198"},{"key":"ref60","doi-asserted-by":"publisher","DOI":"10.1145\/2987550.2987560"},{"key":"ref61","first-page":"541","article-title":"SimAI: Unifying architecture design and performance tuning for large-scale large language model training with scalability and precision","volume-title":"Proc. 22nd USENIX Symp. Networked Syst. Design Implement.","author":"Wang"},{"key":"ref62","article-title":"Qwen3 technical report","volume-title":"arXiv:2505.09388","author":"Yang","year":"2025"},{"key":"ref63","doi-asserted-by":"publisher","DOI":"10.1287\/mnsc.17.11.712"},{"key":"ref64","doi-asserted-by":"publisher","DOI":"10.1145\/2342356.2342390"},{"key":"ref65","doi-asserted-by":"publisher","DOI":"10.1145\/3098822.3098841"},{"key":"ref66","doi-asserted-by":"publisher","DOI":"10.1145\/3359989.3365426"},{"key":"ref67","volume-title":"Deepep: An Efficient Expert-Parallel Communication Library","author":"Zhao et al","year":"2025"},{"key":"ref68","doi-asserted-by":"publisher","DOI":"10.1145\/2592798.2592803"},{"key":"ref69","article-title":"An extensible software transport layer for GPU networking","author":"Zhou","year":"2025","journal-title":"arXiv:2504.17307"}],"container-title":["IEEE Transactions on Networking"],"original-title":[],"link":[{"URL":"http:\/\/xplorestaging.ieee.org\/ielx8\/10723154\/11317935\/11450478.pdf?arnumber=11450478","content-type":"unspecified","content-version":"vor","intended-application":"similarity-checking"}],"deposited":{"date-parts":[[2026,4,1]],"date-time":"2026-04-01T20:11:49Z","timestamp":1775074309000},"score":1,"resource":{"primary":{"URL":"https:\/\/ieeexplore.ieee.org\/document\/11450478\/"}},"subtitle":[],"short-title":[],"issued":{"date-parts":[[2026]]},"references-count":69,"URL":"https:\/\/doi.org\/10.1109\/ton.2026.3676382","relation":{},"ISSN":["2998-4157"],"issn-type":[{"value":"2998-4157","type":"electronic"}],"subject":[],"published":{"date-parts":[[2026]]}}}