{"id":"https://openalex.org/W4415482672","doi":"https://doi.org/10.1109/tcc.2025.3624752","title":"Castor: Optimizing Deep Learning Job Scheduling in Multi-Tenant GPU Clusters via Intelligent Colocation","display_name":"Castor: Optimizing Deep Learning Job Scheduling in Multi-Tenant GPU Clusters via Intelligent Colocation","publication_year":2025,"publication_date":"2025-10-23","ids":{"openalex":"https://openalex.org/W4415482672","doi":"https://doi.org/10.1109/tcc.2025.3624752"},"language":null,"primary_location":{"id":"doi:10.1109/tcc.2025.3624752","is_oa":false,"landing_page_url":"https://doi.org/10.1109/tcc.2025.3624752","pdf_url":null,"source":{"id":"https://openalex.org/S2492498579","display_name":"IEEE Transactions on Cloud Computing","issn_l":"2168-7161","issn":["2168-7161","2372-0018"],"is_oa":false,"is_in_doaj":false,"is_core":true,"host_organization":"https://openalex.org/P4310319808","host_organization_name":"Institute of Electrical and Electronics Engineers","host_organization_lineage":["https://openalex.org/P4310319808"],"host_organization_lineage_names":["Institute of Electrical and Electronics Engineers"],"type":"journal"},"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"IEEE Transactions on Cloud Computing","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":null,"display_name":"Yizhou Luo","orcid":"https://orcid.org/0009-0002-0331-4278"},"institutions":[{"id":"https://openalex.org/I204983213","display_name":"Harbin Institute of Technology","ror":"https://ror.org/01yqg2h08","country_code":"CN","type":"education","lineage":["https://openalex.org/I204983213"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Yizhou Luo","raw_affiliation_strings":["Harbin Institute of Technology, Shenzhen, China"],"raw_orcid":"https://orcid.org/0009-0002-0331-4278","affiliations":[{"raw_affiliation_string":"Harbin Institute of Technology, Shenzhen, China","institution_ids":["https://openalex.org/I204983213"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5103044683","display_name":"Jiaxin Lai","orcid":"https://orcid.org/0009-0003-5272-2283"},"institutions":[{"id":"https://openalex.org/I204983213","display_name":"Harbin Institute of Technology","ror":"https://ror.org/01yqg2h08","country_code":"CN","type":"education","lineage":["https://openalex.org/I204983213"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Jiaxin Lai","raw_affiliation_strings":["Harbin Institute of Technology, Shenzhen, China"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Harbin Institute of Technology, Shenzhen, China","institution_ids":["https://openalex.org/I204983213"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5016836702","display_name":"Shaohuai Shi","orcid":"https://orcid.org/0000-0002-1418-5160"},"institutions":[{"id":"https://openalex.org/I204983213","display_name":"Harbin Institute of Technology","ror":"https://ror.org/01yqg2h08","country_code":"CN","type":"education","lineage":["https://openalex.org/I204983213"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Shaohuai Shi","raw_affiliation_strings":["Harbin Institute of Technology, Shenzhen, China"],"raw_orcid":"https://orcid.org/0000-0002-1418-5160","affiliations":[{"raw_affiliation_string":"Harbin Institute of Technology, Shenzhen, China","institution_ids":["https://openalex.org/I204983213"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5100418393","display_name":"Chen Chen","orcid":"https://orcid.org/0000-0001-9480-5632"},"institutions":[{"id":"https://openalex.org/I183067930","display_name":"Shanghai Jiao Tong University","ror":"https://ror.org/0220qvk04","country_code":"CN","type":"education","lineage":["https://openalex.org/I183067930"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Chen Chen","raw_affiliation_strings":["John Hopcroft Center for Computer Science, Shanghai Jiao Tong University, Shanghai, China","John Hopcroft Center for Computer Science, Shanghai Jiao Tong University, China"],"raw_orcid":"https://orcid.org/0000-0001-9480-5632","affiliations":[{"raw_affiliation_string":"John Hopcroft Center for Computer Science, Shanghai Jiao Tong University, Shanghai, China","institution_ids":["https://openalex.org/I183067930"]},{"raw_affiliation_string":"John Hopcroft Center for Computer Science, Shanghai Jiao Tong University, China","institution_ids":["https://openalex.org/I183067930"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5050911130","display_name":"Shuhan Qi","orcid":"https://orcid.org/0000-0002-6903-145X"},"institutions":[{"id":"https://openalex.org/I204983213","display_name":"Harbin Institute of Technology","ror":"https://ror.org/01yqg2h08","country_code":"CN","type":"education","lineage":["https://openalex.org/I204983213"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Shuhan Qi","raw_affiliation_strings":["Harbin Institute of Technology, Shenzhen, China"],"raw_orcid":"https://orcid.org/0000-0002-6903-145X","affiliations":[{"raw_affiliation_string":"Harbin Institute of Technology, Shenzhen, China","institution_ids":["https://openalex.org/I204983213"]}]},{"author_position":"middle","author":{"id":null,"display_name":"Jiajia Zhang","orcid":"https://orcid.org/0000-0001-6611-2046"},"institutions":[{"id":"https://openalex.org/I204983213","display_name":"Harbin Institute of Technology","ror":"https://ror.org/01yqg2h08","country_code":"CN","type":"education","lineage":["https://openalex.org/I204983213"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Jiajia Zhang","raw_affiliation_strings":["Harbin Institute of Technology, Shenzhen, China"],"raw_orcid":"https://orcid.org/0000-0001-6611-2046","affiliations":[{"raw_affiliation_string":"Harbin Institute of Technology, Shenzhen, China","institution_ids":["https://openalex.org/I204983213"]}]},{"author_position":"last","author":{"id":null,"display_name":"Qiang Wang","orcid":"https://orcid.org/0000-0002-2986-967X"},"institutions":[{"id":"https://openalex.org/I204983213","display_name":"Harbin Institute of Technology","ror":"https://ror.org/01yqg2h08","country_code":"CN","type":"education","lineage":["https://openalex.org/I204983213"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Qiang Wang","raw_affiliation_strings":["Harbin Institute of Technology, Shenzhen, China"],"raw_orcid":"https://orcid.org/0000-0002-2986-967X","affiliations":[{"raw_affiliation_string":"Harbin Institute of Technology, Shenzhen, China","institution_ids":["https://openalex.org/I204983213"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":2,"corresponding_author_ids":[],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":0.0,"has_fulltext":false,"cited_by_count":0,"citation_normalized_percentile":{"value":0.29423394,"is_in_top_1_percent":false,"is_in_top_10_percent":false},"cited_by_percentile_year":null,"biblio":{"volume":"14","issue":"1","first_page":"45","last_page":"61"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T10054","display_name":"Parallel Computing and Optimization Techniques","score":0.902999997138977,"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":0.902999997138977,"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"}}],"keywords":[{"id":"https://openalex.org/keywords/workload","display_name":"Workload","score":0.5649999976158142},{"id":"https://openalex.org/keywords/job-shop-scheduling","display_name":"Job shop scheduling","score":0.5515000224113464},{"id":"https://openalex.org/keywords/scheduling","display_name":"Scheduling (production processes)","score":0.5494999885559082},{"id":"https://openalex.org/keywords/gpu-cluster","display_name":"GPU cluster","score":0.49970000982284546},{"id":"https://openalex.org/keywords/cloud-computing","display_name":"Cloud computing","score":0.42160001397132874},{"id":"https://openalex.org/keywords/deep-learning","display_name":"Deep learning","score":0.39719998836517334},{"id":"https://openalex.org/keywords/rate-monotonic-scheduling","display_name":"Rate-monotonic scheduling","score":0.3919999897480011},{"id":"https://openalex.org/keywords/software-deployment","display_name":"Software deployment","score":0.3865000009536743},{"id":"https://openalex.org/keywords/queueing-theory","display_name":"Queueing theory","score":0.37700000405311584},{"id":"https://openalex.org/keywords/flow-shop-scheduling","display_name":"Flow shop scheduling","score":0.36500000953674316}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.8628000020980835},{"id":"https://openalex.org/C120314980","wikidata":"https://www.wikidata.org/wiki/Q180634","display_name":"Distributed computing","level":1,"score":0.6011999845504761},{"id":"https://openalex.org/C2778476105","wikidata":"https://www.wikidata.org/wiki/Q628539","display_name":"Workload","level":2,"score":0.5649999976158142},{"id":"https://openalex.org/C55416958","wikidata":"https://www.wikidata.org/wiki/Q6206757","display_name":"Job shop scheduling","level":3,"score":0.5515000224113464},{"id":"https://openalex.org/C206729178","wikidata":"https://www.wikidata.org/wiki/Q2271896","display_name":"Scheduling (production processes)","level":2,"score":0.5494999885559082},{"id":"https://openalex.org/C2781335571","wikidata":"https://www.wikidata.org/wiki/Q2633544","display_name":"GPU cluster","level":3,"score":0.49970000982284546},{"id":"https://openalex.org/C79974875","wikidata":"https://www.wikidata.org/wiki/Q483639","display_name":"Cloud computing","level":2,"score":0.42160001397132874},{"id":"https://openalex.org/C108583219","wikidata":"https://www.wikidata.org/wiki/Q197536","display_name":"Deep learning","level":2,"score":0.39719998836517334},{"id":"https://openalex.org/C127456818","wikidata":"https://www.wikidata.org/wiki/Q238879","display_name":"Rate-monotonic scheduling","level":4,"score":0.3919999897480011},{"id":"https://openalex.org/C105339364","wikidata":"https://www.wikidata.org/wiki/Q2297740","display_name":"Software deployment","level":2,"score":0.3865000009536743},{"id":"https://openalex.org/C22684755","wikidata":"https://www.wikidata.org/wiki/Q847526","display_name":"Queueing theory","level":2,"score":0.37700000405311584},{"id":"https://openalex.org/C158336966","wikidata":"https://www.wikidata.org/wiki/Q3074426","display_name":"Flow shop scheduling","level":4,"score":0.36500000953674316},{"id":"https://openalex.org/C111873713","wikidata":"https://www.wikidata.org/wiki/Q1641413","display_name":"Job scheduler","level":3,"score":0.35260000824928284},{"id":"https://openalex.org/C31689143","wikidata":"https://www.wikidata.org/wiki/Q733809","display_name":"Fair-share scheduling","level":3,"score":0.3506999909877777},{"id":"https://openalex.org/C107568181","wikidata":"https://www.wikidata.org/wiki/Q5319000","display_name":"Dynamic priority scheduling","level":3,"score":0.33889999985694885},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.32260000705718994},{"id":"https://openalex.org/C51332947","wikidata":"https://www.wikidata.org/wiki/Q1172305","display_name":"Shared resource","level":2,"score":0.31690001487731934},{"id":"https://openalex.org/C19012869","wikidata":"https://www.wikidata.org/wiki/Q578372","display_name":"Response time","level":2,"score":0.3140999972820282},{"id":"https://openalex.org/C57869625","wikidata":"https://www.wikidata.org/wiki/Q1783502","display_name":"Rate of convergence","level":3,"score":0.30059999227523804},{"id":"https://openalex.org/C202491316","wikidata":"https://www.wikidata.org/wiki/Q272683","display_name":"Instruction set","level":2,"score":0.2921999990940094},{"id":"https://openalex.org/C206952183","wikidata":"https://www.wikidata.org/wiki/Q1193100","display_name":"Preemption","level":2,"score":0.28780001401901245},{"id":"https://openalex.org/C79403827","wikidata":"https://www.wikidata.org/wiki/Q3988","display_name":"Real-time computing","level":1,"score":0.2833000123500824},{"id":"https://openalex.org/C119948110","wikidata":"https://www.wikidata.org/wiki/Q7858726","display_name":"Two-level scheduling","level":4,"score":0.27799999713897705},{"id":"https://openalex.org/C2779851693","wikidata":"https://www.wikidata.org/wiki/Q183484","display_name":"Graphics processing unit","level":2,"score":0.27709999680519104},{"id":"https://openalex.org/C175154964","wikidata":"https://www.wikidata.org/wiki/Q380077","display_name":"Task analysis","level":3,"score":0.2727000117301941},{"id":"https://openalex.org/C173608175","wikidata":"https://www.wikidata.org/wiki/Q232661","display_name":"Parallel computing","level":1,"score":0.27160000801086426},{"id":"https://openalex.org/C48044578","wikidata":"https://www.wikidata.org/wiki/Q727490","display_name":"Scalability","level":2,"score":0.2703999876976013},{"id":"https://openalex.org/C68339613","wikidata":"https://www.wikidata.org/wiki/Q1549489","display_name":"Speedup","level":2,"score":0.267300009727478},{"id":"https://openalex.org/C164866538","wikidata":"https://www.wikidata.org/wiki/Q367351","display_name":"Cluster (spacecraft)","level":2,"score":0.2606000006198883},{"id":"https://openalex.org/C115537543","wikidata":"https://www.wikidata.org/wiki/Q165596","display_name":"Cache","level":2,"score":0.2599000036716461},{"id":"https://openalex.org/C50630238","wikidata":"https://www.wikidata.org/wiki/Q971505","display_name":"General-purpose computing on graphics processing units","level":3,"score":0.2590999901294708},{"id":"https://openalex.org/C132439773","wikidata":"https://www.wikidata.org/wiki/Q1810747","display_name":"Least slack time scheduling","level":5,"score":0.2590000033378601},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.258899986743927}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.1109/tcc.2025.3624752","is_oa":false,"landing_page_url":"https://doi.org/10.1109/tcc.2025.3624752","pdf_url":null,"source":{"id":"https://openalex.org/S2492498579","display_name":"IEEE Transactions on Cloud Computing","issn_l":"2168-7161","issn":["2168-7161","2372-0018"],"is_oa":false,"is_in_doaj":false,"is_core":true,"host_organization":"https://openalex.org/P4310319808","host_organization_name":"Institute of Electrical and Electronics Engineers","host_organization_lineage":["https://openalex.org/P4310319808"],"host_organization_lineage_names":["Institute of Electrical and Electronics Engineers"],"type":"journal"},"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"IEEE Transactions on Cloud Computing","raw_type":"journal-article"}],"best_oa_location":null,"sustainable_development_goals":[],"awards":[{"id":"https://openalex.org/G2989850567","display_name":null,"funder_award_id":"62302126","funder_id":"https://openalex.org/F4320321001","funder_display_name":"National Natural Science Foundation of China"},{"id":"https://openalex.org/G6150324345","display_name":null,"funder_award_id":"62302123","funder_id":"https://openalex.org/F4320321001","funder_display_name":"National Natural Science Foundation of China"}],"funders":[{"id":"https://openalex.org/F4320321001","display_name":"National Natural Science Foundation of China","ror":"https://ror.org/01h0zpd94"}],"has_content":{"pdf":false,"grobid_xml":false},"content_urls":null,"referenced_works_count":30,"referenced_works":["https://openalex.org/W1512808804","https://openalex.org/W1572016165","https://openalex.org/W2034008996","https://openalex.org/W2074988668","https://openalex.org/W2131613942","https://openalex.org/W2335814492","https://openalex.org/W2760837370","https://openalex.org/W2798515322","https://openalex.org/W2918843973","https://openalex.org/W2919115771","https://openalex.org/W2919897868","https://openalex.org/W2920534947","https://openalex.org/W2963149401","https://openalex.org/W2969388332","https://openalex.org/W2982339048","https://openalex.org/W2982346115","https://openalex.org/W2991040477","https://openalex.org/W3012479151","https://openalex.org/W3197816522","https://openalex.org/W4221149636","https://openalex.org/W4288071417","https://openalex.org/W4297882618","https://openalex.org/W4318541676","https://openalex.org/W4383749535","https://openalex.org/W4387321109","https://openalex.org/W4393159654","https://openalex.org/W4394923413","https://openalex.org/W4395020669","https://openalex.org/W4402896827","https://openalex.org/W4408941332"],"related_works":[],"abstract_inverted_index":{"Deep":[0],"learning":[1],"(DL)":[2],"has":[3],"achieved":[4],"significant":[5],"success":[6],"across":[7,200],"a":[8,97,118,142,191,223],"wide":[9],"range":[10],"of":[11,17,57,74,83,131,184,193],"domains,":[12],"prompting":[13],"the":[14,72,81,128,182,234],"widespread":[15],"deployment":[16],"GPU":[18,58,94,151,163],"clusters":[19,42],"equipped":[20],"with":[21,71],"specialized":[22],"accelerators":[23],"to":[24,99,126,148,175,215,242],"support":[25],"high-performance":[26],"training":[27,137],"workloads.":[28],"To":[29],"minimize":[30],"operational":[31],"costs":[32],"while":[33,180],"maximizing":[34],"resource":[35,101],"utilization,":[36],"efficient":[37,145],"job":[38,105,119,160,209],"scheduling":[39,120,146,172],"in":[40,108],"these":[41,79,114],"is":[43],"essential.":[44],"Although":[45],"recent":[46],"schedulers":[47],"have":[48,91],"improved":[49],"cluster":[50],"efficiency":[51],"through":[52,187],"periodic":[53],"reallocation":[54],"or":[55],"selection":[56],"resources,":[59],"they":[60],"still":[61],"face":[62],"challenges":[63],"such":[64],"as":[65,96],"preemption":[66],"and":[67,103,165,171,197],"migration":[68],"overheads,":[69],"along":[70],"risk":[73],"degrading":[75],"model":[76,121],"accuracy.":[77],"Despite":[78],"limitations,":[80],"potential":[82],"GPUsharing":[84],"remains":[85],"largely":[86],"underexplored.":[87],"Few":[88],"existing":[89],"studies":[90],"systematically":[92],"examined":[93],"sharing":[95,164,237],"strategy":[98],"enhance":[100],"utilization":[102],"reduce":[104],"queuing":[106],"delays":[107],"multi-tenant":[109],"DL":[110,155,185,195,218],"clusters.":[111],"Motivated":[112],"by":[113,212,240],"insights,":[115],"we":[116,203],"propose":[117],"that":[122,205],"enables":[123],"multiple":[124,154],"jobs":[125],"share":[127],"same":[129],"set":[130],"GPUs":[132],"without":[133],"modifying":[134],"their":[135],"original":[136],"configurations.":[138],"We":[139],"introduce":[140],"Castor,":[141],"simple":[143],"yet":[144],"system,":[147],"achieve":[149],"intelligent":[150],"colocation":[152],"for":[153,162],"jobs.":[156],"Castor":[157,206,227],"intelligently":[158],"selects":[159],"pairs":[161],"determines":[166],"runtime":[167],"parameters":[168],"(sub-batch":[169],"size":[170],"time":[173,211],"point)":[174],"optimize":[176],"overall":[177],"system":[178],"performance":[179],"preserving":[181],"accuracy":[183],"convergence":[186],"gradient":[188],"accumulation.":[189],"Through":[190],"combination":[192],"physical":[194],"workloads":[196],"trace-driven":[198],"simulations":[199],"various":[201],"configurations,":[202,232],"demonstrate":[204],"reduces":[207],"average":[208],"completion":[210],"26\u201352%":[213],"compared":[214],"state-of-the-art":[216],"preemptive":[217],"schedulers,":[219],"despite":[220],"operating":[221],"under":[222],"preemption-free":[224],"policy.":[225],"Furthermore,":[226],"effectively":[228],"identifies":[229],"optimal":[230],"resource-sharing":[231],"outperforming":[233],"baseline":[235],"first-fit":[236],"policy":[238],"(SJF-FFS)":[239],"up":[241],"20%":[243],"on":[244],"large-scale":[245],"workload":[246],"traces.":[247]},"counts_by_year":[],"updated_date":"2026-03-27T05:58:40.876381","created_date":"2025-10-23T00:00:00"}