{"id":"https://openalex.org/W4414629895","doi":"https://doi.org/10.1145/3680207.3723493","title":"D2MoE: Dual Routing and Dynamic Scheduling for Efficient On-Device MoE-based LLM Serving","display_name":"D2MoE: Dual Routing and Dynamic Scheduling for Efficient On-Device MoE-based LLM Serving","publication_year":2025,"publication_date":"2025-11-03","ids":{"openalex":"https://openalex.org/W4414629895","doi":"https://doi.org/10.1145/3680207.3723493"},"language":"en","primary_location":{"id":"doi:10.1145/3680207.3723493","is_oa":true,"landing_page_url":"https://doi.org/10.1145/3680207.3723493","pdf_url":"https://dl.acm.org/doi/pdf/10.1145/3680207.3723493","source":null,"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Proceedings of the 31st Annual International Conference on Mobile Computing and Networking","raw_type":"proceedings-article"},"type":"conference-paper","indexed_in":["arxiv","crossref"],"open_access":{"is_oa":true,"oa_status":"gold","oa_url":"https://dl.acm.org/doi/pdf/10.1145/3680207.3723493","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":null,"display_name":"Haodong Wang","orcid":"https://orcid.org/0009-0001-8977-850X"},"institutions":[{"id":"https://openalex.org/I200769079","display_name":"Hong Kong University of Science and Technology","ror":"https://ror.org/00q4vv597","country_code":"HK","type":"education","lineage":["https://openalex.org/I200769079"]}],"countries":["HK"],"is_corresponding":false,"raw_author_name":"Haodong Wang","raw_affiliation_strings":["Hong Kong University of Science and Technology, Hong Kong, Hong Kong"],"raw_orcid":"https://orcid.org/0009-0001-8977-850X","affiliations":[{"raw_affiliation_string":"Hong Kong University of Science and Technology, Hong Kong, Hong Kong","institution_ids":["https://openalex.org/I200769079"]}]},{"author_position":"middle","author":{"id":null,"display_name":"Qihua Zhou","orcid":"https://orcid.org/0000-0003-0328-2894"},"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"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Qihua Zhou","raw_affiliation_strings":["College of Computer Science and Software Engineering, Shenzhen University, Shenzhen, China"],"raw_orcid":"https://orcid.org/0000-0003-0328-2894","affiliations":[{"raw_affiliation_string":"College of Computer Science and Software Engineering, Shenzhen University, Shenzhen, China","institution_ids":["https://openalex.org/I180726961"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5075795671","display_name":"Zicong Hong","orcid":"https://orcid.org/0000-0001-5689-382X"},"institutions":[{"id":"https://openalex.org/I200769079","display_name":"Hong Kong University of Science and Technology","ror":"https://ror.org/00q4vv597","country_code":"HK","type":"education","lineage":["https://openalex.org/I200769079"]}],"countries":["HK"],"is_corresponding":false,"raw_author_name":"Zicong Hong","raw_affiliation_strings":["Hong Kong University of Science and Technology, Hong Kong, Hong Kong"],"raw_orcid":"https://orcid.org/0000-0001-5689-382X","affiliations":[{"raw_affiliation_string":"Hong Kong University of Science and Technology, Hong Kong, Hong Kong","institution_ids":["https://openalex.org/I200769079"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5043464306","display_name":"Song Guo","orcid":"https://orcid.org/0000-0001-9831-2202"},"institutions":[{"id":"https://openalex.org/I200769079","display_name":"Hong Kong University of Science and Technology","ror":"https://ror.org/00q4vv597","country_code":"HK","type":"education","lineage":["https://openalex.org/I200769079"]}],"countries":["HK"],"is_corresponding":false,"raw_author_name":"Song Guo","raw_affiliation_strings":["Hong Kong University of Science and Technology, Hong Kong, Hong Kong"],"raw_orcid":"https://orcid.org/0000-0001-9831-2202","affiliations":[{"raw_affiliation_string":"Hong Kong University of Science and Technology, Hong Kong, Hong Kong","institution_ids":["https://openalex.org/I200769079"]}]}],"institutions":[],"countries_distinct_count":2,"institutions_distinct_count":2,"corresponding_author_ids":[],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":null,"has_fulltext":true,"cited_by_count":2,"citation_normalized_percentile":null,"cited_by_percentile_year":null,"biblio":{"volume":null,"issue":null,"first_page":"574","last_page":"588"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T10273","display_name":"IoT and Edge/Fog Computing","score":0.983299970626831,"subfield":{"id":"https://openalex.org/subfields/1705","display_name":"Computer Networks and Communications"},"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/T10273","display_name":"IoT and Edge/Fog Computing","score":0.983299970626831,"subfield":{"id":"https://openalex.org/subfields/1705","display_name":"Computer Networks and Communications"},"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/T10080","display_name":"Energy Efficient Wireless Sensor Networks","score":0.9758999943733215,"subfield":{"id":"https://openalex.org/subfields/1705","display_name":"Computer Networks and Communications"},"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/T11932","display_name":"Wireless Body Area Networks","score":0.9545999765396118,"subfield":{"id":"https://openalex.org/subfields/2204","display_name":"Biomedical Engineering"},"field":{"id":"https://openalex.org/fields/22","display_name":"Engineering"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}}],"keywords":[{"id":"https://openalex.org/keywords/scheduling","display_name":"Scheduling (production processes)","score":0.5925999879837036},{"id":"https://openalex.org/keywords/memory-footprint","display_name":"Memory footprint","score":0.5479000210762024},{"id":"https://openalex.org/keywords/inference","display_name":"Inference","score":0.5281000137329102},{"id":"https://openalex.org/keywords/computation","display_name":"Computation","score":0.46970000863075256},{"id":"https://openalex.org/keywords/edge-device","display_name":"Edge device","score":0.4366999864578247},{"id":"https://openalex.org/keywords/queue","display_name":"Queue","score":0.4092999994754791},{"id":"https://openalex.org/keywords/cloud-computing","display_name":"Cloud computing","score":0.3702999949455261},{"id":"https://openalex.org/keywords/inference-engine","display_name":"Inference engine","score":0.3682999908924103},{"id":"https://openalex.org/keywords/heuristics","display_name":"Heuristics","score":0.36169999837875366},{"id":"https://openalex.org/keywords/enhanced-data-rates-for-gsm-evolution","display_name":"Enhanced Data Rates for GSM Evolution","score":0.36070001125335693}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.8572999835014343},{"id":"https://openalex.org/C206729178","wikidata":"https://www.wikidata.org/wiki/Q2271896","display_name":"Scheduling (production processes)","level":2,"score":0.5925999879837036},{"id":"https://openalex.org/C74912251","wikidata":"https://www.wikidata.org/wiki/Q6815727","display_name":"Memory footprint","level":2,"score":0.5479000210762024},{"id":"https://openalex.org/C2776214188","wikidata":"https://www.wikidata.org/wiki/Q408386","display_name":"Inference","level":2,"score":0.5281000137329102},{"id":"https://openalex.org/C120314980","wikidata":"https://www.wikidata.org/wiki/Q180634","display_name":"Distributed computing","level":1,"score":0.47609999775886536},{"id":"https://openalex.org/C45374587","wikidata":"https://www.wikidata.org/wiki/Q12525525","display_name":"Computation","level":2,"score":0.46970000863075256},{"id":"https://openalex.org/C138236772","wikidata":"https://www.wikidata.org/wiki/Q25098575","display_name":"Edge device","level":3,"score":0.4366999864578247},{"id":"https://openalex.org/C160403385","wikidata":"https://www.wikidata.org/wiki/Q220543","display_name":"Queue","level":2,"score":0.4092999994754791},{"id":"https://openalex.org/C79974875","wikidata":"https://www.wikidata.org/wiki/Q483639","display_name":"Cloud computing","level":2,"score":0.3702999949455261},{"id":"https://openalex.org/C46743427","wikidata":"https://www.wikidata.org/wiki/Q1341685","display_name":"Inference engine","level":3,"score":0.3682999908924103},{"id":"https://openalex.org/C127705205","wikidata":"https://www.wikidata.org/wiki/Q5748245","display_name":"Heuristics","level":2,"score":0.36169999837875366},{"id":"https://openalex.org/C162307627","wikidata":"https://www.wikidata.org/wiki/Q204833","display_name":"Enhanced Data Rates for GSM Evolution","level":2,"score":0.36070001125335693},{"id":"https://openalex.org/C115537543","wikidata":"https://www.wikidata.org/wiki/Q165596","display_name":"Cache","level":2,"score":0.3587000072002411},{"id":"https://openalex.org/C28855332","wikidata":"https://www.wikidata.org/wiki/Q198099","display_name":"Quantization (signal processing)","level":2,"score":0.3522000014781952},{"id":"https://openalex.org/C108010975","wikidata":"https://www.wikidata.org/wiki/Q500094","display_name":"Pruning","level":2,"score":0.3490000069141388},{"id":"https://openalex.org/C173801870","wikidata":"https://www.wikidata.org/wiki/Q201413","display_name":"Heuristic","level":2,"score":0.33970001339912415},{"id":"https://openalex.org/C43521106","wikidata":"https://www.wikidata.org/wiki/Q2165493","display_name":"Pipeline (software)","level":2,"score":0.3368000090122223},{"id":"https://openalex.org/C142614401","wikidata":"https://www.wikidata.org/wiki/Q777433","display_name":"Forward chaining","level":3,"score":0.326200008392334},{"id":"https://openalex.org/C26324664","wikidata":"https://www.wikidata.org/wiki/Q1065525","display_name":"Message queue","level":2,"score":0.3122999966144562},{"id":"https://openalex.org/C2780451532","wikidata":"https://www.wikidata.org/wiki/Q759676","display_name":"Task (project management)","level":2,"score":0.3098999857902527},{"id":"https://openalex.org/C172191483","wikidata":"https://www.wikidata.org/wiki/Q1071806","display_name":"Provisioning","level":2,"score":0.29980000853538513},{"id":"https://openalex.org/C157764524","wikidata":"https://www.wikidata.org/wiki/Q1383412","display_name":"Throughput","level":3,"score":0.29580000042915344},{"id":"https://openalex.org/C162262903","wikidata":"https://www.wikidata.org/wiki/Q343527","display_name":"Allocator","level":2,"score":0.29440000653266907},{"id":"https://openalex.org/C74172769","wikidata":"https://www.wikidata.org/wiki/Q1446839","display_name":"Routing (electronic design automation)","level":2,"score":0.29109999537467957},{"id":"https://openalex.org/C79403827","wikidata":"https://www.wikidata.org/wiki/Q3988","display_name":"Real-time computing","level":1,"score":0.28790000081062317},{"id":"https://openalex.org/C2778456923","wikidata":"https://www.wikidata.org/wiki/Q5337692","display_name":"Edge computing","level":3,"score":0.28630000352859497},{"id":"https://openalex.org/C22684755","wikidata":"https://www.wikidata.org/wiki/Q847526","display_name":"Queueing theory","level":2,"score":0.2766999900341034},{"id":"https://openalex.org/C124304363","wikidata":"https://www.wikidata.org/wiki/Q673661","display_name":"Abstraction","level":2,"score":0.274399995803833},{"id":"https://openalex.org/C107568181","wikidata":"https://www.wikidata.org/wiki/Q5319000","display_name":"Dynamic priority scheduling","level":3,"score":0.2615000009536743},{"id":"https://openalex.org/C7366592","wikidata":"https://www.wikidata.org/wiki/Q1255620","display_name":"Dram","level":2,"score":0.25589999556541443}],"mesh":[],"locations_count":3,"locations":[{"id":"doi:10.1145/3680207.3723493","is_oa":true,"landing_page_url":"https://doi.org/10.1145/3680207.3723493","pdf_url":"https://dl.acm.org/doi/pdf/10.1145/3680207.3723493","source":null,"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Proceedings of the 31st Annual International Conference on Mobile Computing and Networking","raw_type":"proceedings-article"},{"id":"pmh:oai:arXiv.org:2504.15299","is_oa":true,"landing_page_url":"http://arxiv.org/abs/2504.15299","pdf_url":"https://arxiv.org/pdf/2504.15299","source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":null,"issn":null,"is_oa":true,"is_in_doaj":false,"is_core":false,"host_organization":"https://openalex.org/I205783295","host_organization_name":"Cornell University","host_organization_lineage":["https://openalex.org/I205783295"],"host_organization_lineage_names":[],"type":"repository"},"license":null,"license_id":null,"version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},{"id":"pmh:oai:repository.hkust.edu.hk:1783.1-168844","is_oa":false,"landing_page_url":"http://repository.hkust.edu.hk/ir/Record/1783.1-168844","pdf_url":null,"source":{"id":"https://openalex.org/S4306401796","display_name":"Rare & Special e-Zone (The Hong Kong University of Science and Technology)","issn_l":null,"issn":null,"is_oa":false,"is_in_doaj":false,"is_core":false,"host_organization":"https://openalex.org/I200769079","host_organization_name":"Hong Kong University of Science and Technology","host_organization_lineage":["https://openalex.org/I200769079"],"host_organization_lineage_names":[],"type":"repository"},"license":null,"license_id":null,"version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"Conference paper"}],"best_oa_location":{"id":"doi:10.1145/3680207.3723493","is_oa":true,"landing_page_url":"https://doi.org/10.1145/3680207.3723493","pdf_url":"https://dl.acm.org/doi/pdf/10.1145/3680207.3723493","source":null,"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Proceedings of the 31st Annual International Conference on Mobile Computing and Networking","raw_type":"proceedings-article"},"sustainable_development_goals":[],"awards":[{"id":"https://openalex.org/G3802291057","display_name":null,"funder_award_id":"R5060-19","funder_id":"https://openalex.org/F4320307285","funder_display_name":"Impact Fund"}],"funders":[{"id":"https://openalex.org/F4320307285","display_name":"Impact Fund","ror":"https://ror.org/00jb20j87"},{"id":"https://openalex.org/F4320321001","display_name":"National Natural Science Foundation of China","ror":"https://ror.org/01h0zpd94"},{"id":"https://openalex.org/F4320323537","display_name":"Hong Kong University of Science and Technology","ror":"https://ror.org/00q4vv597"}],"has_content":{"grobid_xml":true,"pdf":true},"content_urls":{"pdf":"https://content.openalex.org/works/W4414629895.pdf","grobid_xml":"https://content.openalex.org/works/W4414629895.grobid-xml"},"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"The":[0],"mixture":[1],"of":[2,10,45,93,128,155],"experts":[3,198],"(MoE)":[4],"model":[5,54,73],"is":[6,31],"a":[7,18,144,165],"sparse":[8],"variant":[9],"large":[11],"language":[12],"models":[13],"(LLMs),":[14],"designed":[15],"to":[16,35,63,69,99,119,138,199,216,226],"hold":[17],"better":[19],"balance":[20],"between":[21,179],"intelligent":[22],"capability":[23],"and":[24,61,147,163,181,218],"computational":[25],"overhead.":[26],"Despite":[27],"its":[28,241],"benefits,":[29],"MoE":[30,65],"still":[32,235],"too":[33],"expensive":[34],"deploy":[36],"on":[37,202],"resource-constrained":[38],"edge":[39,204],"devices,":[40],"especially":[41],"with":[42],"the":[43,80,90,101,115,125,133,149,160,176,191,197,210,220,229],"demands":[44],"on-device":[46,91,231],"inference":[47,212,232],"services.":[48],"Recent":[49],"research":[50],"efforts":[51],"often":[52],"apply":[53],"compression":[55],"techniques,":[56],"such":[57],"as":[58,240],"quantization,":[59],"pruning":[60],"merging,":[62],"restrict":[64],"complexity.":[66],"Unfortunately,":[67],"due":[68],"their":[70],"predefined":[71],"static":[72],"optimization":[74],"strategies,":[75],"they":[76],"cannot":[77],"always":[78],"achieve":[79],"desired":[81],"quality-overhead":[82],"trade-off":[83],"when":[84],"handling":[85],"multiple":[86],"requests,":[87],"finally":[88],"degrading":[89],"quality":[92],"service.":[94],"These":[95],"limitations":[96],"motivate":[97],"us":[98],"propose":[100,132],"D2MoE,":[102],"an":[103],"algorithm-system":[104],"co-design":[105],"framework":[106],"that":[107,207],"matches":[108],"diverse":[109],"task":[110],"requirements":[111],"by":[112,124,214,224],"dynamically":[113],"allocating":[114],"most":[116],"proper":[117],"bit-width":[118],"each":[120],"expert.":[121],"Specifically,":[122],"inspired":[123],"nested":[126],"structure":[127],"matryoshka":[129,134],"dolls,":[130],"we":[131,157],"weight":[135],"quantization":[136],"(MWQ)":[137],"progressively":[139],"compress":[140],"expert":[141,177],"weights":[142],"in":[143],"bit-nested":[145],"manner":[146],"reduce":[148],"required":[150],"runtime":[151],"memory.":[152],"On":[153],"top":[154],"it,":[156],"further":[158],"optimize":[159],"I/O-computation":[161],"pipeline":[162],"design":[164],"heuristic":[166],"scheduling":[167],"algorithm":[168],"following":[169],"our":[170],"hottest-expert-bit-first":[171],"(HEBF)":[172],"principle,":[173],"which":[174],"maximizes":[175],"parallelism":[178],"I/O":[180],"computation":[182],"queue":[183],"under":[184],"constrained":[185],"memory":[186,222],"budgets,":[187],"thus":[188],"significantly":[189],"reducing":[190],"idle":[192],"temporal":[193],"bubbles":[194],"waiting":[195],"for":[196],"load.":[200],"Evaluations":[201],"real":[203],"devices":[205],"show":[206],"D2MoE":[208],"improves":[209],"overall":[211],"throughput":[213],"up":[215,225],"1.39\u00d7":[217],"reduces":[219],"peak":[221],"footprint":[223],"53%":[227],"over":[228],"latest":[230],"frameworks,":[233],"while":[234],"preserving":[236],"comparable":[237],"serving":[238],"accuracy":[239],"INT8":[242],"counterparts.":[243]},"counts_by_year":[{"year":2026,"cited_by_count":2}],"updated_date":"2026-07-14T23:27:15.235271","created_date":"2025-10-10T00:00:00"}