{"id":"https://openalex.org/W4409262447","doi":"https://doi.org/10.1109/tc.2025.3558009","title":"TightLLM: Maximizing Throughput for LLM Inference via Adaptive Offloading Policy","display_name":"TightLLM: Maximizing Throughput for LLM Inference via Adaptive Offloading Policy","publication_year":2025,"publication_date":"2025-04-08","ids":{"openalex":"https://openalex.org/W4409262447","doi":"https://doi.org/10.1109/tc.2025.3558009"},"language":"en","primary_location":{"id":"doi:10.1109/tc.2025.3558009","is_oa":false,"landing_page_url":"https://doi.org/10.1109/tc.2025.3558009","pdf_url":null,"source":{"id":"https://openalex.org/S157670870","display_name":"IEEE Transactions on Computers","issn_l":"0018-9340","issn":["0018-9340","1557-9956","2326-3814"],"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 Computers","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/A5102792327","display_name":"Yitao Hu","orcid":"https://orcid.org/0009-0004-0458-0900"},"institutions":[{"id":"https://openalex.org/I162868743","display_name":"Tianjin University","ror":"https://ror.org/012tb2g32","country_code":"CN","type":"education","lineage":["https://openalex.org/I162868743"]}],"countries":["CN"],"is_corresponding":true,"raw_author_name":"Yitao Hu","raw_affiliation_strings":["Tianjin Key Laboratory of Advanced Networking, Department of Intelligence and Computing, Tianjin University, Tianjin, China","Tianjin Key Laboratory of Advanced Networking, Tianjin University, Tianjin, China"],"affiliations":[{"raw_affiliation_string":"Tianjin Key Laboratory of Advanced Networking, Department of Intelligence and Computing, Tianjin University, Tianjin, China","institution_ids":["https://openalex.org/I162868743"]},{"raw_affiliation_string":"Tianjin Key Laboratory of Advanced Networking, Tianjin University, Tianjin, China","institution_ids":["https://openalex.org/I162868743"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5009324770","display_name":"Xiulong Liu","orcid":"https://orcid.org/0000-0002-4746-5599"},"institutions":[{"id":"https://openalex.org/I162868743","display_name":"Tianjin University","ror":"https://ror.org/012tb2g32","country_code":"CN","type":"education","lineage":["https://openalex.org/I162868743"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Xiulong Liu","raw_affiliation_strings":["Tianjin Key Laboratory of Advanced Networking, Department of Intelligence and Computing, Tianjin University, Tianjin, China","Tianjin Key Laboratory of Advanced Networking, Tianjin University, Tianjin, China"],"affiliations":[{"raw_affiliation_string":"Tianjin Key Laboratory of Advanced Networking, Department of Intelligence and Computing, Tianjin University, Tianjin, China","institution_ids":["https://openalex.org/I162868743"]},{"raw_affiliation_string":"Tianjin Key Laboratory of Advanced Networking, Tianjin University, Tianjin, China","institution_ids":["https://openalex.org/I162868743"]}]},{"author_position":"middle","author":{"id":null,"display_name":"Guotao Yang","orcid":"https://orcid.org/0009-0005-9355-9296"},"institutions":[{"id":"https://openalex.org/I162868743","display_name":"Tianjin University","ror":"https://ror.org/012tb2g32","country_code":"CN","type":"education","lineage":["https://openalex.org/I162868743"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Guotao Yang","raw_affiliation_strings":["Tianjin Key Laboratory of Advanced Networking, Department of Intelligence and Computing, Tianjin University, Tianjin, China","Tianjin Key Laboratory of Advanced Networking, Tianjin University, Tianjin, China"],"affiliations":[{"raw_affiliation_string":"Tianjin Key Laboratory of Advanced Networking, Department of Intelligence and Computing, Tianjin University, Tianjin, China","institution_ids":["https://openalex.org/I162868743"]},{"raw_affiliation_string":"Tianjin Key Laboratory of Advanced Networking, Tianjin University, Tianjin, China","institution_ids":["https://openalex.org/I162868743"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5117075958","display_name":"Linxuan Li","orcid":null},"institutions":[{"id":"https://openalex.org/I162868743","display_name":"Tianjin University","ror":"https://ror.org/012tb2g32","country_code":"CN","type":"education","lineage":["https://openalex.org/I162868743"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Linxuan Li","raw_affiliation_strings":["Tianjin Key Laboratory of Advanced Networking, Department of Intelligence and Computing, Tianjin University, Tianjin, China","Tianjin Key Laboratory of Advanced Networking, Tianjin University, Tianjin, China"],"affiliations":[{"raw_affiliation_string":"Tianjin Key Laboratory of Advanced Networking, Department of Intelligence and Computing, Tianjin University, Tianjin, China","institution_ids":["https://openalex.org/I162868743"]},{"raw_affiliation_string":"Tianjin Key Laboratory of Advanced Networking, Tianjin University, Tianjin, China","institution_ids":["https://openalex.org/I162868743"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5117075959","display_name":"Kai Zeng","orcid":null},"institutions":[{"id":"https://openalex.org/I162868743","display_name":"Tianjin University","ror":"https://ror.org/012tb2g32","country_code":"CN","type":"education","lineage":["https://openalex.org/I162868743"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Kai Zeng","raw_affiliation_strings":["Tianjin Key Laboratory of Advanced Networking, Department of Intelligence and Computing, Tianjin University, Tianjin, China","Tianjin Key Laboratory of Advanced Networking, Tianjin University, Tianjin, China"],"affiliations":[{"raw_affiliation_string":"Tianjin Key Laboratory of Advanced Networking, Department of Intelligence and Computing, Tianjin University, Tianjin, China","institution_ids":["https://openalex.org/I162868743"]},{"raw_affiliation_string":"Tianjin Key Laboratory of Advanced Networking, Tianjin University, Tianjin, China","institution_ids":["https://openalex.org/I162868743"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5101695421","display_name":"Zhixin Zhao","orcid":"https://orcid.org/0009-0009-7865-4905"},"institutions":[{"id":"https://openalex.org/I162868743","display_name":"Tianjin University","ror":"https://ror.org/012tb2g32","country_code":"CN","type":"education","lineage":["https://openalex.org/I162868743"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Zhixin Zhao","raw_affiliation_strings":["Tianjin Key Laboratory of Advanced Networking, Department of Intelligence and Computing, Tianjin University, Tianjin, China","Tianjin Key Laboratory of Advanced Networking, Tianjin University, Tianjin, China"],"affiliations":[{"raw_affiliation_string":"Tianjin Key Laboratory of Advanced Networking, Department of Intelligence and Computing, Tianjin University, Tianjin, China","institution_ids":["https://openalex.org/I162868743"]},{"raw_affiliation_string":"Tianjin Key Laboratory of Advanced Networking, Tianjin University, Tianjin, China","institution_ids":["https://openalex.org/I162868743"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5100320971","display_name":"Sheng Chen","orcid":"https://orcid.org/0000-0001-7038-4407"},"institutions":[{"id":"https://openalex.org/I162868743","display_name":"Tianjin University","ror":"https://ror.org/012tb2g32","country_code":"CN","type":"education","lineage":["https://openalex.org/I162868743"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Sheng Chen","raw_affiliation_strings":["Tianjin Key Laboratory of Advanced Networking, Department of Intelligence and Computing, Tianjin University, Tianjin, China","Tianjin Key Laboratory of Advanced Networking, Tianjin University, Tianjin, China"],"affiliations":[{"raw_affiliation_string":"Tianjin Key Laboratory of Advanced Networking, Department of Intelligence and Computing, Tianjin University, Tianjin, China","institution_ids":["https://openalex.org/I162868743"]},{"raw_affiliation_string":"Tianjin Key Laboratory of Advanced Networking, Tianjin University, Tianjin, China","institution_ids":["https://openalex.org/I162868743"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5050753595","display_name":"Laiping Zhao","orcid":"https://orcid.org/0000-0003-1967-2192"},"institutions":[{"id":"https://openalex.org/I162868743","display_name":"Tianjin University","ror":"https://ror.org/012tb2g32","country_code":"CN","type":"education","lineage":["https://openalex.org/I162868743"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Laiping Zhao","raw_affiliation_strings":["Tianjin Key Laboratory of Advanced Networking, Department of Intelligence and Computing, Tianjin University, Tianjin, China","Tianjin Key Laboratory of Advanced Networking, Tianjin University, Tianjin, China"],"affiliations":[{"raw_affiliation_string":"Tianjin Key Laboratory of Advanced Networking, Department of Intelligence and Computing, Tianjin University, Tianjin, China","institution_ids":["https://openalex.org/I162868743"]},{"raw_affiliation_string":"Tianjin Key Laboratory of Advanced Networking, Tianjin University, Tianjin, China","institution_ids":["https://openalex.org/I162868743"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5100397205","display_name":"Wenxin Li","orcid":"https://orcid.org/0000-0001-8507-0339"},"institutions":[{"id":"https://openalex.org/I162868743","display_name":"Tianjin University","ror":"https://ror.org/012tb2g32","country_code":"CN","type":"education","lineage":["https://openalex.org/I162868743"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Wenxin Li","raw_affiliation_strings":["Tianjin Key Laboratory of Advanced Networking, Department of Intelligence and Computing, Tianjin University, Tianjin, China","Tianjin Key Laboratory of Advanced Networking, Tianjin University, Tianjin, China"],"affiliations":[{"raw_affiliation_string":"Tianjin Key Laboratory of Advanced Networking, Department of Intelligence and Computing, Tianjin University, Tianjin, China","institution_ids":["https://openalex.org/I162868743"]},{"raw_affiliation_string":"Tianjin Key Laboratory of Advanced Networking, Tianjin University, Tianjin, China","institution_ids":["https://openalex.org/I162868743"]}]},{"author_position":"last","author":{"id":null,"display_name":"Keqiu Li","orcid":"https://orcid.org/0000-0003-1758-3030"},"institutions":[{"id":"https://openalex.org/I162868743","display_name":"Tianjin University","ror":"https://ror.org/012tb2g32","country_code":"CN","type":"education","lineage":["https://openalex.org/I162868743"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Keqiu Li","raw_affiliation_strings":["Tianjin Key Laboratory of Advanced Networking, Department of Intelligence and Computing, Tianjin University, Tianjin, China","Tianjin Key Laboratory of Advanced Networking, Tianjin University, Tianjin, China"],"affiliations":[{"raw_affiliation_string":"Tianjin Key Laboratory of Advanced Networking, Department of Intelligence and Computing, Tianjin University, Tianjin, China","institution_ids":["https://openalex.org/I162868743"]},{"raw_affiliation_string":"Tianjin Key Laboratory of Advanced Networking, Tianjin University, Tianjin, China","institution_ids":["https://openalex.org/I162868743"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":10,"corresponding_author_ids":["https://openalex.org/A5102792327"],"corresponding_institution_ids":["https://openalex.org/I162868743"],"apc_list":null,"apc_paid":null,"fwci":3.9769,"has_fulltext":false,"cited_by_count":5,"citation_normalized_percentile":{"value":0.93331946,"is_in_top_1_percent":false,"is_in_top_10_percent":true},"cited_by_percentile_year":{"min":95,"max":100},"biblio":{"volume":"74","issue":"7","first_page":"2195","last_page":"2209"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T12702","display_name":"Brain Tumor Detection and Classification","score":0.9818000197410583,"subfield":{"id":"https://openalex.org/subfields/2808","display_name":"Neurology"},"field":{"id":"https://openalex.org/fields/28","display_name":"Neuroscience"},"domain":{"id":"https://openalex.org/domains/1","display_name":"Life Sciences"}},"topics":[{"id":"https://openalex.org/T12702","display_name":"Brain Tumor Detection and Classification","score":0.9818000197410583,"subfield":{"id":"https://openalex.org/subfields/2808","display_name":"Neurology"},"field":{"id":"https://openalex.org/fields/28","display_name":"Neuroscience"},"domain":{"id":"https://openalex.org/domains/1","display_name":"Life Sciences"}},{"id":"https://openalex.org/T10764","display_name":"Privacy-Preserving Technologies in Data","score":0.9632999897003174,"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/T11612","display_name":"Stochastic Gradient Optimization Techniques","score":0.9035999774932861,"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"}}],"keywords":[{"id":"https://openalex.org/keywords/throughput","display_name":"Throughput","score":0.7329528331756592},{"id":"https://openalex.org/keywords/computer-science","display_name":"Computer science","score":0.7237640619277954},{"id":"https://openalex.org/keywords/inference","display_name":"Inference","score":0.6255790591239929},{"id":"https://openalex.org/keywords/computer-network","display_name":"Computer network","score":0.3290966749191284},{"id":"https://openalex.org/keywords/artificial-intelligence","display_name":"Artificial intelligence","score":0.2717241942882538},{"id":"https://openalex.org/keywords/wireless","display_name":"Wireless","score":0.15950971841812134},{"id":"https://openalex.org/keywords/telecommunications","display_name":"Telecommunications","score":0.11039263010025024}],"concepts":[{"id":"https://openalex.org/C157764524","wikidata":"https://www.wikidata.org/wiki/Q1383412","display_name":"Throughput","level":3,"score":0.7329528331756592},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.7237640619277954},{"id":"https://openalex.org/C2776214188","wikidata":"https://www.wikidata.org/wiki/Q408386","display_name":"Inference","level":2,"score":0.6255790591239929},{"id":"https://openalex.org/C31258907","wikidata":"https://www.wikidata.org/wiki/Q1301371","display_name":"Computer network","level":1,"score":0.3290966749191284},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.2717241942882538},{"id":"https://openalex.org/C555944384","wikidata":"https://www.wikidata.org/wiki/Q249","display_name":"Wireless","level":2,"score":0.15950971841812134},{"id":"https://openalex.org/C76155785","wikidata":"https://www.wikidata.org/wiki/Q418","display_name":"Telecommunications","level":1,"score":0.11039263010025024}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.1109/tc.2025.3558009","is_oa":false,"landing_page_url":"https://doi.org/10.1109/tc.2025.3558009","pdf_url":null,"source":{"id":"https://openalex.org/S157670870","display_name":"IEEE Transactions on Computers","issn_l":"0018-9340","issn":["0018-9340","1557-9956","2326-3814"],"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 Computers","raw_type":"journal-article"}],"best_oa_location":null,"sustainable_development_goals":[],"awards":[{"id":"https://openalex.org/G3404210196","display_name":null,"funder_award_id":"62202328","funder_id":"https://openalex.org/F4320321001","funder_display_name":"National Natural Science Foundation of China"},{"id":"https://openalex.org/G6975830235","display_name":null,"funder_award_id":"2022YFB4501000","funder_id":"https://openalex.org/F4320335777","funder_display_name":"National Key Research and Development Program of China"}],"funders":[{"id":"https://openalex.org/F4320321001","display_name":"National Natural Science Foundation of China","ror":"https://ror.org/01h0zpd94"},{"id":"https://openalex.org/F4320335777","display_name":"National Key Research and Development Program of China","ror":null}],"has_content":{"grobid_xml":false,"pdf":false},"content_urls":null,"referenced_works_count":34,"referenced_works":["https://openalex.org/W3033428961","https://openalex.org/W3081168214","https://openalex.org/W3083934060","https://openalex.org/W3091048326","https://openalex.org/W3092288641","https://openalex.org/W3213380720","https://openalex.org/W4296473245","https://openalex.org/W4296557505","https://openalex.org/W4385992398","https://openalex.org/W4387321091","https://openalex.org/W4391853740","https://openalex.org/W4392353733","https://openalex.org/W4394841768","https://openalex.org/W4398173942","https://openalex.org/W4402671659","https://openalex.org/W4403337153","https://openalex.org/W4403937333","https://openalex.org/W4404401017","https://openalex.org/W4404401018","https://openalex.org/W4408844835","https://openalex.org/W6772383348","https://openalex.org/W6790051600","https://openalex.org/W6811340617","https://openalex.org/W6811928498","https://openalex.org/W6838322825","https://openalex.org/W6838633097","https://openalex.org/W6849514112","https://openalex.org/W6849563483","https://openalex.org/W6850625674","https://openalex.org/W6850927664","https://openalex.org/W6856619137","https://openalex.org/W6856884471","https://openalex.org/W6857690716","https://openalex.org/W6869966162"],"related_works":["https://openalex.org/W4391375266","https://openalex.org/W2899084033","https://openalex.org/W2748952813","https://openalex.org/W2390279801","https://openalex.org/W4391913857","https://openalex.org/W2358668433","https://openalex.org/W4396701345","https://openalex.org/W2376932109","https://openalex.org/W2001405890","https://openalex.org/W4396696052"],"abstract_inverted_index":{"Large":[0],"language":[1],"models":[2,240],"(LLMs)":[3],"have":[4],"demonstrated":[5],"remarkable":[6],"performance":[7],"across":[8,91,149,232],"a":[9,85,180,295],"wide":[10],"range":[11],"of":[12,192,235,298],"tasks,":[13,289],"largely":[14],"due":[15],"to":[16,51,96,107,115,126,144,152,183,246,257,267,272,302],"their":[17],"substantial":[18,296],"model":[19,211],"size.":[20],"However,":[21,102],"this":[22,157],"also":[23],"results":[24],"in":[25,262,276],"significant":[26],"GPU":[27,40,56,65,82,128,204,236],"memory":[28,57],"demands":[29],"during":[30,121,251],"inference.":[31],"To":[32,72,156],"address":[33,184],"these":[34],"challenges":[35],"on":[36],"hardware":[37,237],"with":[38,81,170,200],"limited":[39],"memory,":[41,53],"existing":[42],"approaches":[43],"employ":[44],"offloading":[45,60,104,139,269],"techniques":[46],"that":[47,138,242],"offload":[48],"unused":[49],"tensors":[50],"CPU":[52],"thereby":[54],"reducing":[55],"usage.":[58],"Since":[59],"involves":[61],"data":[62,79,198],"transfer":[63,70,80,119,147,186,199],"between":[64],"and":[66,130,161,202,213,227,238,255,278,293],"CPU,":[67],"it":[68],"introduces":[69],"overhead.":[71],"mitigate":[73],"this,":[74],"prior":[75],"works":[76],"typically":[77],"overlap":[78],"computation":[83,201],"using":[84],"fixed":[86],"pipelining":[87],"strategy":[88,182],"applied":[89],"uniformly":[90],"all":[92],"inference":[93,109,123,132,150,154,166],"iterations,":[94],"referred":[95],"as":[97],"<italic":[98],"xmlns:mml=\"http://www.w3.org/1998/Math/MathML\"":[99,135],"xmlns:xlink=\"http://www.w3.org/1999/xlink\">static</i>":[100],"offloading.":[101],"static":[103],"policies":[105,140],"fail":[106],"maximize":[108,153],"throughput":[110,250,261,275],"because":[111],"they":[112],"cannot":[113],"adapt":[114],"the":[116,122,145,193,215,222,252,263,273,284],"dynamically":[117,189],"changing":[118],"overhead":[120,148,187,226],"process,":[124],"leading":[125],"increasing":[127],"idleness":[129],"reduced":[131],"throughput.":[133,155,230],"<p":[134],"xmlns:xlink=\"http://www.w3.org/1999/xlink\">We":[136],"propose":[137],"should":[141],"be":[142],"<i>adaptive</i>":[143],"varying":[146],"iterations":[151],"end,":[158],"we":[159],"design":[160],"implement":[162],"an":[163],"adaptive":[164],"offloading-based":[165],"system":[167],"called":[168],"TightLLM":[169,243,281],"two":[171],"key":[172],"innovations.":[173],"First,":[174],"its":[175],"key-value":[176],"(KV)":[177],"distributor":[178],"employs":[179],"<i>trade-compute-for-transfer</i>":[181],"growing":[185],"by":[188,300],"recomputing":[190],"portions":[191],"KV":[194],"cache,":[195],"effectively":[196],"overlapping":[197],"minimizing":[203],"idleness.":[205],"Second,":[206],"TightLLM\u2019s":[207],"weight":[208,224],"loader":[209],"slices":[210],"weights":[212],"distributes":[214],"loading":[216,225],"process":[217],"<i>across":[218],"multiple":[219],"batches</i>,":[220],"amortizing":[221],"excessive":[223],"significantly":[228],"improving":[229],"Evaluation":[231],"various":[233],"combinations":[234],"LLM":[239],"shows":[241],"achieves":[244],"1.3":[245],"23":[247],"times":[248,259],"higher":[249,260,274],"decoding":[253,279],"phase":[254,265],"1.2":[256],"22":[258],"prefill":[264,277],"compared":[266],"state-of-the-art":[268],"systems.":[270],"Due":[271],"phases,":[280],"can":[282],"reduce":[283],"completion":[285],"time":[286],"for":[287],"large-scale":[288],"which":[290],"involve":[291],"processing":[292],"generating":[294],"number":[297],"tokens,":[299],"59.6%":[301],"94.9%.":[303]},"counts_by_year":[{"year":2026,"cited_by_count":3},{"year":2025,"cited_by_count":2}],"updated_date":"2026-04-09T08:11:56.329763","created_date":"2025-10-10T00:00:00"}
