{"id":"https://openalex.org/W4417471602","doi":"https://doi.org/10.1109/ton.2025.3645806","title":"Optimizing Mixture-of-Experts Inference Time via Model Deployment and Communication Scheduling","display_name":"Optimizing Mixture-of-Experts Inference Time via Model Deployment and Communication Scheduling","publication_year":2025,"publication_date":"2025-12-18","ids":{"openalex":"https://openalex.org/W4417471602","doi":"https://doi.org/10.1109/ton.2025.3645806"},"language":"en","primary_location":{"id":"doi:10.1109/ton.2025.3645806","is_oa":false,"landing_page_url":"https://doi.org/10.1109/ton.2025.3645806","pdf_url":null,"source":{"id":"https://openalex.org/S5407042750","display_name":"IEEE Transactions on Networking","issn_l":"2998-4157","issn":["2998-4157"],"is_oa":false,"is_in_doaj":false,"is_core":false,"host_organization":null,"host_organization_name":null,"host_organization_lineage":[],"host_organization_lineage_names":[],"type":"journal"},"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"IEEE Transactions on Networking","raw_type":"journal-article"},"type":"article","indexed_in":["crossref"],"open_access":{"is_oa":false,"oa_status":"closed","oa_url":null,"any_repository_has_fulltext":false},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5100708117","display_name":"Jialong Li","orcid":"https://orcid.org/0000-0003-3416-5551"},"institutions":[{"id":"https://openalex.org/I180726961","display_name":"Shenzhen University","ror":"https://ror.org/01vy4gh70","country_code":"CN","type":"education","lineage":["https://openalex.org/I180726961"]},{"id":"https://openalex.org/I4210152380","display_name":"Shenzhen Technology University","ror":"https://ror.org/04qzpec27","country_code":"CN","type":"education","lineage":["https://openalex.org/I4210152380"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Jialong Li","raw_affiliation_strings":["Shenzhen University of Advanced Technology, Shenzhen, China","Shenzhen University of Advanced Technology, China"],"raw_orcid":"https://orcid.org/0000-0003-3416-5551","affiliations":[{"raw_affiliation_string":"Shenzhen University of Advanced Technology, Shenzhen, China","institution_ids":["https://openalex.org/I4210152380","https://openalex.org/I180726961"]},{"raw_affiliation_string":"Shenzhen University of Advanced Technology, China","institution_ids":["https://openalex.org/I4210152380","https://openalex.org/I180726961"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5025000935","display_name":"Shreyansh Tripathi","orcid":"https://orcid.org/0000-0001-9518-7557"},"institutions":[{"id":"https://openalex.org/I4210109712","display_name":"Max Planck Institute for Informatics","ror":"https://ror.org/01w19ak89","country_code":"DE","type":"facility","lineage":["https://openalex.org/I149899117","https://openalex.org/I4210109712"]}],"countries":["DE"],"is_corresponding":false,"raw_author_name":"Shreyansh Tripathi","raw_affiliation_strings":["Max Planck Institute for Informatics, Saarbrücken, Germany","Max Planck Institute for Informatics, Germany"],"raw_orcid":"https://orcid.org/0000-0001-9518-7557","affiliations":[{"raw_affiliation_string":"Max Planck Institute for Informatics, Saarbrücken, Germany","institution_ids":["https://openalex.org/I4210109712"]},{"raw_affiliation_string":"Max Planck Institute for Informatics, Germany","institution_ids":["https://openalex.org/I4210109712"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5099111088","display_name":"Lakshay Rastogi","orcid":null},"institutions":[{"id":"https://openalex.org/I4210109712","display_name":"Max Planck Institute for Informatics","ror":"https://ror.org/01w19ak89","country_code":"DE","type":"facility","lineage":["https://openalex.org/I149899117","https://openalex.org/I4210109712"]}],"countries":["DE"],"is_corresponding":false,"raw_author_name":"Lakshay Rastogi","raw_affiliation_strings":["Max Planck Institute for Informatics, Saarbrücken, Germany","Max Planck Institute for Informatics, Germany"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Max Planck Institute for Informatics, Saarbrücken, Germany","institution_ids":["https://openalex.org/I4210109712"]},{"raw_affiliation_string":"Max Planck Institute for Informatics, Germany","institution_ids":["https://openalex.org/I4210109712"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5101774629","display_name":"Yiming Lei","orcid":"https://orcid.org/0000-0002-2158-547X"},"institutions":[{"id":"https://openalex.org/I4210109712","display_name":"Max Planck Institute for Informatics","ror":"https://ror.org/01w19ak89","country_code":"DE","type":"facility","lineage":["https://openalex.org/I149899117","https://openalex.org/I4210109712"]}],"countries":["DE"],"is_corresponding":false,"raw_author_name":"Yiming Lei","raw_affiliation_strings":["Max Planck Institute for Informatics, Saarbrücken, Germany","Max Planck Institute for Informatics, Germany"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Max Planck Institute for Informatics, Saarbrücken, Germany","institution_ids":["https://openalex.org/I4210109712"]},{"raw_affiliation_string":"Max Planck Institute for Informatics, Germany","institution_ids":["https://openalex.org/I4210109712"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5100638027","display_name":"Rui Pan","orcid":"https://orcid.org/0000-0002-6973-3259"},"institutions":[{"id":"https://openalex.org/I20089843","display_name":"Princeton University","ror":"https://ror.org/00hx57361","country_code":"US","type":"education","lineage":["https://openalex.org/I20089843"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Rui Pan","raw_affiliation_strings":["Princeton University, Princeton, USA","Princeton University, USA"],"raw_orcid":"https://orcid.org/0000-0002-6973-3259","affiliations":[{"raw_affiliation_string":"Princeton University, Princeton, USA","institution_ids":["https://openalex.org/I20089843"]},{"raw_affiliation_string":"Princeton University, USA","institution_ids":["https://openalex.org/I20089843"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5037947752","display_name":"Yiting Xia","orcid":"https://orcid.org/0000-0003-2981-1459"},"institutions":[{"id":"https://openalex.org/I4210109712","display_name":"Max Planck Institute for Informatics","ror":"https://ror.org/01w19ak89","country_code":"DE","type":"facility","lineage":["https://openalex.org/I149899117","https://openalex.org/I4210109712"]}],"countries":["DE"],"is_corresponding":false,"raw_author_name":"Yiting Xia","raw_affiliation_strings":["Max Planck Institute for Informatics, Saarbrücken, Germany","Max Planck Institute for Informatics, Germany"],"raw_orcid":"https://orcid.org/0000-0003-2981-1459","affiliations":[{"raw_affiliation_string":"Max Planck Institute for Informatics, Saarbrücken, Germany","institution_ids":["https://openalex.org/I4210109712"]},{"raw_affiliation_string":"Max Planck Institute for Informatics, Germany","institution_ids":["https://openalex.org/I4210109712"]}]}],"institutions":[],"countries_distinct_count":3,"institutions_distinct_count":6,"corresponding_author_ids":[],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":1.7588,"has_fulltext":false,"cited_by_count":1,"citation_normalized_percentile":{"value":0.89936517,"is_in_top_1_percent":false,"is_in_top_10_percent":false},"cited_by_percentile_year":{"min":91,"max":95},"biblio":{"volume":"34","issue":null,"first_page":"2478","last_page":"2497"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T10764","display_name":"Privacy-Preserving Technologies in Data","score":0.1746000051498413,"subfield":{"id":"https://openalex.org/subfields/1702","display_name":"Artificial Intelligence"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},"topics":[{"id":"https://openalex.org/T10764","display_name":"Privacy-Preserving Technologies in Data","score":0.1746000051498413,"subfield":{"id":"https://openalex.org/subfields/1702","display_name":"Artificial Intelligence"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},{"id":"https://openalex.org/T11704","display_name":"Mobile Crowdsensing and Crowdsourcing","score":0.15729999542236328,"subfield":{"id":"https://openalex.org/subfields/1706","display_name":"Computer Science Applications"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},{"id":"https://openalex.org/T10036","display_name":"Advanced Neural Network Applications","score":0.1453000009059906,"subfield":{"id":"https://openalex.org/subfields/1707","display_name":"Computer Vision and Pattern Recognition"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}}],"keywords":[{"id":"https://openalex.org/keywords/scheduling","display_name":"Scheduling (production processes)","score":0.714900016784668},{"id":"https://openalex.org/keywords/software-deployment","display_name":"Software deployment","score":0.7050999999046326},{"id":"https://openalex.org/keywords/comet","display_name":"Comet","score":0.6287999749183655},{"id":"https://openalex.org/keywords/inference","display_name":"Inference","score":0.531499981880188},{"id":"https://openalex.org/keywords/homogeneous","display_name":"Homogeneous","score":0.49729999899864197},{"id":"https://openalex.org/keywords/overhead","display_name":"Overhead (engineering)","score":0.4544999897480011}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.7907999753952026},{"id":"https://openalex.org/C206729178","wikidata":"https://www.wikidata.org/wiki/Q2271896","display_name":"Scheduling (production processes)","level":2,"score":0.714900016784668},{"id":"https://openalex.org/C105339364","wikidata":"https://www.wikidata.org/wiki/Q2297740","display_name":"Software deployment","level":2,"score":0.7050999999046326},{"id":"https://openalex.org/C120314980","wikidata":"https://www.wikidata.org/wiki/Q180634","display_name":"Distributed computing","level":1,"score":0.6424999833106995},{"id":"https://openalex.org/C94081185","wikidata":"https://www.wikidata.org/wiki/Q3559","display_name":"Comet","level":2,"score":0.6287999749183655},{"id":"https://openalex.org/C2776214188","wikidata":"https://www.wikidata.org/wiki/Q408386","display_name":"Inference","level":2,"score":0.531499981880188},{"id":"https://openalex.org/C66882249","wikidata":"https://www.wikidata.org/wiki/Q169336","display_name":"Homogeneous","level":2,"score":0.49729999899864197},{"id":"https://openalex.org/C2779960059","wikidata":"https://www.wikidata.org/wiki/Q7113681","display_name":"Overhead (engineering)","level":2,"score":0.4544999897480011},{"id":"https://openalex.org/C79403827","wikidata":"https://www.wikidata.org/wiki/Q3988","display_name":"Real-time computing","level":1,"score":0.3831000030040741},{"id":"https://openalex.org/C158156997","wikidata":"https://www.wikidata.org/wiki/Q1416645","display_name":"Models of communication","level":2,"score":0.36000001430511475},{"id":"https://openalex.org/C29140674","wikidata":"https://www.wikidata.org/wiki/Q206637","display_name":"Computer cluster","level":2,"score":0.3231000006198883},{"id":"https://openalex.org/C55416958","wikidata":"https://www.wikidata.org/wiki/Q6206757","display_name":"Job shop scheduling","level":3,"score":0.32019999623298645},{"id":"https://openalex.org/C137836250","wikidata":"https://www.wikidata.org/wiki/Q984063","display_name":"Optimization problem","level":2,"score":0.296999990940094},{"id":"https://openalex.org/C2780451532","wikidata":"https://www.wikidata.org/wiki/Q759676","display_name":"Task (project management)","level":2,"score":0.26570001244544983},{"id":"https://openalex.org/C93996380","wikidata":"https://www.wikidata.org/wiki/Q44127","display_name":"Server","level":2,"score":0.2639000117778778}],"mesh":[],"locations_count":2,"locations":[{"id":"doi:10.1109/ton.2025.3645806","is_oa":false,"landing_page_url":"https://doi.org/10.1109/ton.2025.3645806","pdf_url":null,"source":{"id":"https://openalex.org/S5407042750","display_name":"IEEE Transactions on Networking","issn_l":"2998-4157","issn":["2998-4157"],"is_oa":false,"is_in_doaj":false,"is_core":false,"host_organization":null,"host_organization_name":null,"host_organization_lineage":[],"host_organization_lineage_names":[],"type":"journal"},"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"IEEE Transactions on Networking","raw_type":"journal-article"},{"id":"pmh:oai:pure.mpg.de:item_3685744","is_oa":false,"landing_page_url":"http://hdl.handle.net/21.11116/0000-0012-512C-0","pdf_url":null,"source":{"id":"https://openalex.org/S4306400654","display_name":"MPG.PuRe (Max Planck Society)","issn_l":null,"issn":null,"is_oa":false,"is_in_doaj":false,"is_core":false,"host_organization":"https://openalex.org/I149899117","host_organization_name":"Max Planck Society","host_organization_lineage":["https://openalex.org/I149899117"],"host_organization_lineage_names":[],"type":"repository"},"license":null,"license_id":null,"version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":"IEEE Transactions on Networking","raw_type":"info:eu-repo/semantics/article"}],"best_oa_location":null,"sustainable_development_goals":[],"awards":[],"funders":[],"has_content":{"grobid_xml":false,"pdf":false},"content_urls":null,"referenced_works_count":24,"referenced_works":["https://openalex.org/W977523618","https://openalex.org/W2018346931","https://openalex.org/W2065841482","https://openalex.org/W2149804187","https://openalex.org/W2157529519","https://openalex.org/W2530088943","https://openalex.org/W2744387122","https://openalex.org/W2794896506","https://openalex.org/W2798515322","https://openalex.org/W3086105743","https://openalex.org/W3106803824","https://openalex.org/W3129831491","https://openalex.org/W4220967350","https://openalex.org/W4226057994","https://openalex.org/W4283397908","https://openalex.org/W4290991419","https://openalex.org/W4308995136","https://openalex.org/W4364382874","https://openalex.org/W4386396242","https://openalex.org/W4387321109","https://openalex.org/W4388855670","https://openalex.org/W4394923534","https://openalex.org/W4394946008","https://openalex.org/W4396817294"],"related_works":[],"abstract_inverted_index":{"As":[0],"machine":[1],"learning":[2],"models":[3,18,30,91,117],"scale":[4],"in":[5,66,79,154,191,196],"size":[6],"and":[7,43,58,120,132,172,194],"complexity,":[8],"their":[9],"computational":[10],"requirements":[11],"become":[12],"a":[13,140,146],"significant":[14],"barrier.":[15],"Mixture-of-Experts":[16],"(MoE)":[17],"alleviate":[19],"this":[20],"issue":[21],"by":[22,33,74,86,204],"selectively":[23],"activating":[24],"relevant":[25],"experts.":[26],"Despite":[27],"this,":[28],"MoE":[29,67,165],"are":[31],"hindered":[32],"high":[34],"communication":[35,60,72,173],"overhead":[36],"from":[37,45,89,149],"all-to-all":[38,59,80,100],"operations,":[39],"low":[40],"GPU":[41,47,84,111,202],"utilization,":[42],"complications":[44],"heterogeneous":[46,123,197],"environments.":[48,198],"This":[49],"paper":[50],"presents":[51],"Comet,":[52],"which":[53],"optimizes":[54],"both":[55],"model":[56,170],"deployment":[57,171],"scheduling":[61,174],"to":[62,163,189,206,209],"address":[63],"these":[64],"challenges":[65],"inference.":[68],"Comet":[69,125,158,181,200],"achieves":[70],"minimal":[71],"times":[73],"strategically":[75],"ordering":[76],"token":[77],"transmissions":[78],"communications.":[81],"It":[82],"improves":[83],"utilization":[85,203],"colocating":[87],"experts":[88],"different":[90],"on":[92,118],"the":[93,97,134,150,155,160],"same":[94],"device,":[95],"avoiding":[96],"limitations":[98],"of":[99,187],"communication.":[101],"We":[102],"analyze":[103],"Comet\u2019s":[104],"optimization":[105],"strategies":[106],"theoretically":[107],"across":[108,175],"four":[109],"common":[110],"cluster":[112],"settings:":[113],"exclusive":[114],"vs.":[115,122],"colocated":[116],"GPUs,":[119],"homogeneous":[121,192],"GPUs.":[124],"provides":[126],"optimal":[127,169],"solutions":[128],"for":[129,133],"three":[130],"cases,":[131],"remaining":[135],"NP-hard":[136],"scenario,":[137],"it":[138],"offers":[139],"polynomial-time":[141],"sub-optimal":[142],"solution":[143],"with":[144],"only":[145],"1.09\u00d7":[147],"degradation":[148],"optimal,":[151],"as":[152],"shown":[153],"simulation":[156],"results.":[157],"is":[159],"first":[161],"approach":[162],"minimize":[164],"inference":[166],"time":[167],"via":[168],"various":[176],"scenarios.":[177],"Evaluations":[178],"demonstrate":[179],"that":[180],"significantly":[182],"accelerates":[183],"inference,":[184],"achieving":[185],"speedups":[186],"up":[188,205],"2.63\u00d7":[190],"clusters":[193],"2.91\u00d7":[195],"Moreover,":[199],"enhances":[201],"2.38\u00d7":[207],"compared":[208],"existing":[210],"methods.":[211]},"counts_by_year":[{"year":2025,"cited_by_count":1}],"updated_date":"2026-06-11T09:08:48.828518","created_date":"2025-12-18T00:00:00"}