{"id":"https://openalex.org/W7117769016","doi":"https://doi.org/10.1145/3737611.3776613","title":"Transformer-Based Resource and Stage-Aware Scheduling for Model-Parallel LLM Inference","display_name":"Transformer-Based Resource and Stage-Aware Scheduling for Model-Parallel LLM Inference","publication_year":2025,"publication_date":"2025-12-31","ids":{"openalex":"https://openalex.org/W7117769016","doi":"https://doi.org/10.1145/3737611.3776613"},"language":null,"primary_location":{"id":"doi:10.1145/3737611.3776613","is_oa":true,"landing_page_url":"https://doi.org/10.1145/3737611.3776613","pdf_url":null,"source":null,"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Companion Proceedings of the 27th International Conference on Distributed Computing and Networking","raw_type":"proceedings-article"},"type":"conference-paper","indexed_in":["crossref"],"open_access":{"is_oa":true,"oa_status":"gold","oa_url":"https://doi.org/10.1145/3737611.3776613","any_repository_has_fulltext":null},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5121642132","display_name":"Rami Naeem","orcid":null},"institutions":[{"id":"https://openalex.org/I98285908","display_name":"The University of Osaka","ror":"https://ror.org/035t8zc32","country_code":"JP","type":"education","lineage":["https://openalex.org/I98285908"]}],"countries":["JP"],"is_corresponding":false,"raw_author_name":"Rami Naeem","raw_affiliation_strings":["Graduate School of Information Science and Technology, The University of Osaka, Suita, Japan"],"raw_orcid":"https://orcid.org/0009-0007-4329-1935","affiliations":[{"raw_affiliation_string":"Graduate School of Information Science and Technology, The University of Osaka, Suita, Japan","institution_ids":["https://openalex.org/I98285908"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5121622663","display_name":"Tengis Buyantogtokh","orcid":null},"institutions":[{"id":"https://openalex.org/I98285908","display_name":"The University of Osaka","ror":"https://ror.org/035t8zc32","country_code":"JP","type":"education","lineage":["https://openalex.org/I98285908"]}],"countries":["JP"],"is_corresponding":false,"raw_author_name":"Tengis Buyantogtokh","raw_affiliation_strings":["Graduate School of Information Science and Technology, The University of Osaka, Suita, Japan"],"raw_orcid":"https://orcid.org/0009-0005-9560-2091","affiliations":[{"raw_affiliation_string":"Graduate School of Information Science and Technology, The University of Osaka, Suita, Japan","institution_ids":["https://openalex.org/I98285908"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5013317917","display_name":"Hamada Rizk","orcid":"https://orcid.org/0000-0002-8278-8801"},"institutions":[{"id":"https://openalex.org/I98285908","display_name":"The University of Osaka","ror":"https://ror.org/035t8zc32","country_code":"JP","type":"education","lineage":["https://openalex.org/I98285908"]}],"countries":["JP"],"is_corresponding":false,"raw_author_name":"Hamada Rizk","raw_affiliation_strings":["Graduate School of Information Science and Technology, The University of Osaka, Suita, Japan"],"raw_orcid":"https://orcid.org/0000-0002-8278-8801","affiliations":[{"raw_affiliation_string":"Graduate School of Information Science and Technology, The University of Osaka, Suita, Japan","institution_ids":["https://openalex.org/I98285908"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5022543520","display_name":"Tatsuya Amano","orcid":"https://orcid.org/0000-0002-8011-247X"},"institutions":[{"id":"https://openalex.org/I98285908","display_name":"The University of Osaka","ror":"https://ror.org/035t8zc32","country_code":"JP","type":"education","lineage":["https://openalex.org/I98285908"]}],"countries":["JP"],"is_corresponding":false,"raw_author_name":"Tatsuya Amano","raw_affiliation_strings":["Graduate School of Information Science and Technology, The University of Osaka, Suita, Japan"],"raw_orcid":"https://orcid.org/0000-0002-8011-247X","affiliations":[{"raw_affiliation_string":"Graduate School of Information Science and Technology, The University of Osaka, Suita, Japan","institution_ids":["https://openalex.org/I98285908"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5121623277","display_name":"Hirozumi Yamaguchi","orcid":null},"institutions":[{"id":"https://openalex.org/I98285908","display_name":"The University of Osaka","ror":"https://ror.org/035t8zc32","country_code":"JP","type":"education","lineage":["https://openalex.org/I98285908"]}],"countries":["JP"],"is_corresponding":false,"raw_author_name":"Hirozumi Yamaguchi","raw_affiliation_strings":["Graduate School of Information Science and Technology, RIKEN Center for Computational Science, The University of Osaka, Suita, Japan"],"raw_orcid":"https://orcid.org/0000-0003-2273-4876","affiliations":[{"raw_affiliation_string":"Graduate School of Information Science and Technology, RIKEN Center for Computational Science, The University of Osaka, Suita, Japan","institution_ids":["https://openalex.org/I98285908"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":1,"corresponding_author_ids":[],"corresponding_institution_ids":["https://openalex.org/I98285908"],"apc_list":null,"apc_paid":null,"fwci":0.0,"has_fulltext":false,"cited_by_count":0,"citation_normalized_percentile":{"value":0.53755962,"is_in_top_1_percent":false,"is_in_top_10_percent":false},"cited_by_percentile_year":null,"biblio":{"volume":null,"issue":null,"first_page":"72","last_page":"77"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T10054","display_name":"Parallel Computing and Optimization Techniques","score":0.5892999768257141,"subfield":{"id":"https://openalex.org/subfields/1708","display_name":"Hardware and Architecture"},"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/T10054","display_name":"Parallel Computing and Optimization Techniques","score":0.5892999768257141,"subfield":{"id":"https://openalex.org/subfields/1708","display_name":"Hardware and Architecture"},"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/T10904","display_name":"Embedded Systems Design Techniques","score":0.14309999346733093,"subfield":{"id":"https://openalex.org/subfields/1708","display_name":"Hardware and Architecture"},"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/T14347","display_name":"Big Data and Digital Economy","score":0.0763000026345253,"subfield":{"id":"https://openalex.org/subfields/1710","display_name":"Information Systems"},"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":0.6894000172615051},{"id":"https://openalex.org/keywords/bottleneck","display_name":"Bottleneck","score":0.5989999771118164},{"id":"https://openalex.org/keywords/inference","display_name":"Inference","score":0.5260000228881836},{"id":"https://openalex.org/keywords/latency","display_name":"Latency (audio)","score":0.5232999920845032},{"id":"https://openalex.org/keywords/replicate","display_name":"Replicate","score":0.4977000057697296},{"id":"https://openalex.org/keywords/heuristic","display_name":"Heuristic","score":0.45089998841285706}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.8495000004768372},{"id":"https://openalex.org/C206729178","wikidata":"https://www.wikidata.org/wiki/Q2271896","display_name":"Scheduling (production processes)","level":2,"score":0.6894000172615051},{"id":"https://openalex.org/C120314980","wikidata":"https://www.wikidata.org/wiki/Q180634","display_name":"Distributed computing","level":1,"score":0.6265000104904175},{"id":"https://openalex.org/C2780513914","wikidata":"https://www.wikidata.org/wiki/Q18210350","display_name":"Bottleneck","level":2,"score":0.5989999771118164},{"id":"https://openalex.org/C2776214188","wikidata":"https://www.wikidata.org/wiki/Q408386","display_name":"Inference","level":2,"score":0.5260000228881836},{"id":"https://openalex.org/C82876162","wikidata":"https://www.wikidata.org/wiki/Q17096504","display_name":"Latency (audio)","level":2,"score":0.5232999920845032},{"id":"https://openalex.org/C2781162219","wikidata":"https://www.wikidata.org/wiki/Q26250693","display_name":"Replicate","level":2,"score":0.4977000057697296},{"id":"https://openalex.org/C173801870","wikidata":"https://www.wikidata.org/wiki/Q201413","display_name":"Heuristic","level":2,"score":0.45089998841285706},{"id":"https://openalex.org/C66882249","wikidata":"https://www.wikidata.org/wiki/Q169336","display_name":"Homogeneous","level":2,"score":0.4203000068664551},{"id":"https://openalex.org/C123745756","wikidata":"https://www.wikidata.org/wiki/Q1665949","display_name":"Interconnection","level":2,"score":0.3269999921321869},{"id":"https://openalex.org/C55416958","wikidata":"https://www.wikidata.org/wiki/Q6206757","display_name":"Job shop scheduling","level":3,"score":0.3059999942779541},{"id":"https://openalex.org/C2776257435","wikidata":"https://www.wikidata.org/wiki/Q1576430","display_name":"Bandwidth (computing)","level":2,"score":0.2824000120162964},{"id":"https://openalex.org/C31258907","wikidata":"https://www.wikidata.org/wiki/Q1301371","display_name":"Computer network","level":1,"score":0.28200000524520874},{"id":"https://openalex.org/C200157131","wikidata":"https://www.wikidata.org/wiki/Q4854763","display_name":"Bandwidth allocation","level":3,"score":0.2791000008583069},{"id":"https://openalex.org/C158207573","wikidata":"https://www.wikidata.org/wiki/Q5747224","display_name":"Heterogeneous network","level":4,"score":0.2605000138282776}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.1145/3737611.3776613","is_oa":true,"landing_page_url":"https://doi.org/10.1145/3737611.3776613","pdf_url":null,"source":null,"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Companion Proceedings of the 27th International Conference on Distributed Computing and Networking","raw_type":"proceedings-article"}],"best_oa_location":{"id":"doi:10.1145/3737611.3776613","is_oa":true,"landing_page_url":"https://doi.org/10.1145/3737611.3776613","pdf_url":null,"source":null,"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Companion Proceedings of the 27th International Conference on Distributed Computing and Networking","raw_type":"proceedings-article"},"sustainable_development_goals":[],"awards":[],"funders":[],"has_content":{"pdf":false,"grobid_xml":false},"content_urls":null,"referenced_works_count":10,"referenced_works":["https://openalex.org/W2973727699","https://openalex.org/W4289821985","https://openalex.org/W4294782983","https://openalex.org/W4306836169","https://openalex.org/W4386721862","https://openalex.org/W4389761026","https://openalex.org/W4403731508","https://openalex.org/W4404990113","https://openalex.org/W6891916419","https://openalex.org/W7083812211"],"related_works":[],"abstract_inverted_index":{"Current":[0],"large":[1,120],"language":[2],"model":[3],"(LLM)":[4],"serving":[5],"systems":[6],"face":[7],"three":[8,90],"key":[9],"limitations":[10],"in":[11,45],"distributed":[12],"scheduling.":[13],"First,":[14],"most":[15],"parallelization":[16],"strategies":[17],"are":[18,58],"not":[19],"stage-aware:":[20],"they":[21,52,57],"treat":[22],"prefill":[23],"and":[24,33,41,47,75,85,100,109,128,132,138,196],"decode":[25],"as":[26,56,189],"uniform":[27],"workloads":[28],"despite":[29],"their":[30],"distinct":[31],"compute":[32],"communication":[34,83],"profiles.":[35],"Second,":[36],"many":[37],"assume":[38],"homogeneous":[39],"hardware":[40],"ignore":[42],"resource":[43],"diversity":[44],"memory":[46],"bandwidth":[48,188],"across":[49],"nodes.":[50],"Third,":[51],"overlook":[53],"network":[54],"congestion,":[55],"primarily":[59],"designed":[60],"for":[61],"data-center":[62],"environments":[63],"with":[64,72,205],"abundant":[65],"interconnect":[66],"bandwidth.":[67],"We":[68,88],"address":[69],"these":[70],"gaps":[71],"a":[73,92,96,101,124,190],"resource-":[74],"stage-aware":[76],"scheduler":[77,103,145],"that":[78,122,198],"models":[79,121],"heterogeneous":[80,131,155],"GPU":[81],"clusters,":[82,134],"costs,":[84],"per-stage":[86],"characteristics.":[87],"compare":[89],"approaches:":[91],"heuristic":[93,172,203],"stage-based":[94],"policy,":[95],"continuous-batching":[97],"(vLLM-style)":[98],"baseline,":[99],"transformer-based":[102],"trained":[104],"by":[105,148,168],"imitation":[106],"to":[107,150],"replicate":[108],"improve":[110],"the":[111],"heuristic.":[112],"Our":[113],"evaluation":[114],"spans":[115],"eight":[116],"representative":[117],"scenarios":[118],"covering":[119],"exceed":[123],"single":[125],"GPU,":[126],"prefill-dominant":[127],"mixed":[129],"workloads,":[130],"bandwidth-limited":[133],"strict":[135],"SLO":[136],"constraints,":[137],"multi-tenant":[139],"or":[140,154],"elastic":[141],"deployments.":[142],"The":[143],"learned":[144,199],"reduces":[146],"latency":[147,167],"up":[149],"50%":[151],"under":[152],"bandwidth-constrained":[153],"conditions":[156],"while":[157,174],"maintaining":[158],"throughput":[159],"within":[160],"20\u201330%":[161],"of":[162,193],"vLLM.":[163],"It":[164],"further":[165],"improves":[166],"3\u201317%":[169],"over":[170],"its":[171],"teacher":[173],"preserving":[175],"100%":[176],"feasibility.":[177],"Continuous":[178],"batching":[179],"remains":[180],"superior":[181],"on":[182],"high-bandwidth":[183],"fabrics.":[184],"These":[185],"results":[186],"identify":[187],"first-order":[191],"determinant":[192],"optimal":[194],"scheduling":[195],"demonstrate":[197],"schedulers":[200],"can":[201],"unify":[202],"feasibility":[204],"adaptive,":[206],"resource-aware":[207],"optimization.":[208]},"counts_by_year":[],"updated_date":"2026-07-29T14:22:42.915294","created_date":"2025-12-31T00:00:00"}