{"id":"https://openalex.org/W7162446476","doi":"https://doi.org/10.1109/ispass69572.2026.00014","title":"TaxBreak: Unmasking the Hidden Costs of LLM Inference Through Overhead Decomposition","display_name":"TaxBreak: Unmasking the Hidden Costs of LLM Inference Through Overhead Decomposition","publication_year":2026,"publication_date":"2026-04-26","ids":{"openalex":"https://openalex.org/W7162446476","doi":"https://doi.org/10.1109/ispass69572.2026.00014"},"language":null,"primary_location":{"id":"doi:10.1109/ispass69572.2026.00014","is_oa":false,"landing_page_url":"https://doi.org/10.1109/ispass69572.2026.00014","pdf_url":null,"source":null,"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"2026 IEEE International Symposium on Performance Analysis of Systems and Software (ISPASS)","raw_type":"proceedings-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/A5136998084","display_name":"Prabhu Vellaisamy","orcid":null},"institutions":[{"id":"https://openalex.org/I74973139","display_name":"Carnegie Mellon University","ror":"https://ror.org/05x2bcf33","country_code":"US","type":"education","lineage":["https://openalex.org/I74973139"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Prabhu Vellaisamy","raw_affiliation_strings":["Carnegie Mellon University,NeuroAI Computer Architecture Lab (NCAL) Electrical and Computer Engineering Department"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Carnegie Mellon University,NeuroAI Computer Architecture Lab (NCAL) Electrical and Computer Engineering Department","institution_ids":["https://openalex.org/I74973139"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5002195638","display_name":"Shreesh Tripathi","orcid":null},"institutions":[{"id":"https://openalex.org/I74973139","display_name":"Carnegie Mellon University","ror":"https://ror.org/05x2bcf33","country_code":"US","type":"education","lineage":["https://openalex.org/I74973139"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Shreesh Tripathi","raw_affiliation_strings":["Carnegie Mellon University,NeuroAI Computer Architecture Lab (NCAL) Electrical and Computer Engineering Department"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Carnegie Mellon University,NeuroAI Computer Architecture Lab (NCAL) Electrical and Computer Engineering Department","institution_ids":["https://openalex.org/I74973139"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5137084679","display_name":"Vignesh Natarajan","orcid":null},"institutions":[{"id":"https://openalex.org/I74973139","display_name":"Carnegie Mellon University","ror":"https://ror.org/05x2bcf33","country_code":"US","type":"education","lineage":["https://openalex.org/I74973139"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Vignesh Natarajan","raw_affiliation_strings":["Carnegie Mellon University,NeuroAI Computer Architecture Lab (NCAL) Electrical and Computer Engineering Department"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Carnegie Mellon University,NeuroAI Computer Architecture Lab (NCAL) Electrical and Computer Engineering Department","institution_ids":["https://openalex.org/I74973139"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5129522407","display_name":"Surya Santhan Thenarasu","orcid":null},"institutions":[{"id":"https://openalex.org/I74973139","display_name":"Carnegie Mellon University","ror":"https://ror.org/05x2bcf33","country_code":"US","type":"education","lineage":["https://openalex.org/I74973139"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Surya Santhan Thenarasu","raw_affiliation_strings":["Carnegie Mellon University,NeuroAI Computer Architecture Lab (NCAL) Electrical and Computer Engineering Department"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Carnegie Mellon University,NeuroAI Computer Architecture Lab (NCAL) Electrical and Computer Engineering Department","institution_ids":["https://openalex.org/I74973139"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5137065258","display_name":"Shawn Blanton","orcid":null},"institutions":[{"id":"https://openalex.org/I74973139","display_name":"Carnegie Mellon University","ror":"https://ror.org/05x2bcf33","country_code":"US","type":"education","lineage":["https://openalex.org/I74973139"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Shawn Blanton","raw_affiliation_strings":["Carnegie Mellon University,NeuroAI Computer Architecture Lab (NCAL) Electrical and Computer Engineering Department"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Carnegie Mellon University,NeuroAI Computer Architecture Lab (NCAL) Electrical and Computer Engineering Department","institution_ids":["https://openalex.org/I74973139"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5129717438","display_name":"John P. Shen","orcid":null},"institutions":[{"id":"https://openalex.org/I74973139","display_name":"Carnegie Mellon University","ror":"https://ror.org/05x2bcf33","country_code":"US","type":"education","lineage":["https://openalex.org/I74973139"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"John P. Shen","raw_affiliation_strings":["Carnegie Mellon University,NeuroAI Computer Architecture Lab (NCAL) Electrical and Computer Engineering Department"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Carnegie Mellon University,NeuroAI Computer Architecture Lab (NCAL) Electrical and Computer Engineering Department","institution_ids":["https://openalex.org/I74973139"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":6,"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":0.84591653,"is_in_top_1_percent":false,"is_in_top_10_percent":false},"cited_by_percentile_year":null,"biblio":{"volume":null,"issue":null,"first_page":"28","last_page":"38"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T10028","display_name":"Topic Modeling","score":0.09889999777078629,"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/T10028","display_name":"Topic Modeling","score":0.09889999777078629,"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/T10181","display_name":"Natural Language Processing Techniques","score":0.09309999644756317,"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/T13643","display_name":"Artificial Intelligence in Law","score":0.08780000358819962,"subfield":{"id":"https://openalex.org/subfields/3320","display_name":"Political Science and International Relations"},"field":{"id":"https://openalex.org/fields/33","display_name":"Social Sciences"},"domain":{"id":"https://openalex.org/domains/2","display_name":"Social Sciences"}}],"keywords":[{"id":"https://openalex.org/keywords/overhead","display_name":"Overhead (engineering)","score":0.5001999735832214},{"id":"https://openalex.org/keywords/inference","display_name":"Inference","score":0.4781999886035919},{"id":"https://openalex.org/keywords/decomposition","display_name":"Decomposition","score":0.42829999327659607},{"id":"https://openalex.org/keywords/key","display_name":"Key (lock)","score":0.32739999890327454}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.6585999727249146},{"id":"https://openalex.org/C2779960059","wikidata":"https://www.wikidata.org/wiki/Q7113681","display_name":"Overhead (engineering)","level":2,"score":0.5001999735832214},{"id":"https://openalex.org/C2776214188","wikidata":"https://www.wikidata.org/wiki/Q408386","display_name":"Inference","level":2,"score":0.4781999886035919},{"id":"https://openalex.org/C124681953","wikidata":"https://www.wikidata.org/wiki/Q339062","display_name":"Decomposition","level":2,"score":0.42829999327659607},{"id":"https://openalex.org/C124101348","wikidata":"https://www.wikidata.org/wiki/Q172491","display_name":"Data mining","level":1,"score":0.38670000433921814},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.38089999556541443},{"id":"https://openalex.org/C26517878","wikidata":"https://www.wikidata.org/wiki/Q228039","display_name":"Key (lock)","level":2,"score":0.32739999890327454},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.30149999260902405},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.29649999737739563},{"id":"https://openalex.org/C116834253","wikidata":"https://www.wikidata.org/wiki/Q2039217","display_name":"Identification (biology)","level":2,"score":0.2822999954223633},{"id":"https://openalex.org/C79403827","wikidata":"https://www.wikidata.org/wiki/Q3988","display_name":"Real-time computing","level":1,"score":0.26589998602867126}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.1109/ispass69572.2026.00014","is_oa":false,"landing_page_url":"https://doi.org/10.1109/ispass69572.2026.00014","pdf_url":null,"source":null,"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"2026 IEEE International Symposium on Performance Analysis of Systems and Software (ISPASS)","raw_type":"proceedings-article"}],"best_oa_location":null,"sustainable_development_goals":[],"awards":[],"funders":[],"has_content":{"grobid_xml":false,"pdf":false},"content_urls":null,"referenced_works_count":14,"referenced_works":["https://openalex.org/W2130076156","https://openalex.org/W2972574632","https://openalex.org/W3150298648","https://openalex.org/W3157671281","https://openalex.org/W3204998121","https://openalex.org/W4387321091","https://openalex.org/W4389519486","https://openalex.org/W4391993504","https://openalex.org/W4394998532","https://openalex.org/W4412987189","https://openalex.org/W4416004263","https://openalex.org/W7133214447","https://openalex.org/W7133222210","https://openalex.org/W7133227460"],"related_works":[],"abstract_inverted_index":{"Large":[0],"Language":[1],"Model":[2],"(LLM)":[3],"inference":[4,17],"is":[5,41,187],"widely":[6],"used":[7],"in":[8,115],"interactive":[9],"assistants":[10],"and":[11,74,84,87,112,118,178,203],"agentic":[12],"systems.":[13],"In":[14],"latencysensitive":[15],"deployments,":[16],"time":[18],"can":[19,131],"become":[20],"dominated":[21],"by":[22,198,207],"host-side":[23],"overheads.":[24],"Existing":[25],"approaches":[26],"typically":[27],"expose":[28],"this":[29],"cost":[30],"only":[31],"as":[32,223],"an":[33],"aggregate":[34,123],"residual":[35],"or":[36,127,237],"a":[37,56,98,188,191,216,224],"launch/queue":[38],"metric,":[39],"which":[40,46],"often":[42],"insufficient":[43],"to":[44,90,106,209],"identify":[45],"execution":[47,105],"layer":[48],"should":[49,143,232],"be":[50],"optimized.":[51],"This":[52],"work":[53],"presents":[54],"TaxBreak,":[55],"tracedriven":[57],"methodology":[58],"for":[59,180,227],"decomposing":[60],"host-visible":[61,107],"orchestration":[62,196],"overhead":[63,146,197],"into":[64],"three":[65],"components:":[66],"framework":[67],"translation":[68,72],"time,":[69,73],"CUDA":[70],"library":[71],"kernel":[75],"launch-path":[76],"time.":[77],"We":[78,158],"validate":[79],"TaxBreak":[80,137,222],"on":[81],"NVIDIA":[82],"H100":[83],"H200":[85],"systems":[86],"use":[88],"it":[89],"derive":[91],"our":[92],"proposed":[93],"Host-Device":[94],"Balance":[95],"Index":[96],"(HDBI),":[97],"boundedness":[99,128],"summary":[100],"index":[101],"that":[102,122,161,179],"relates":[103],"device-active":[104],"orchestration.":[108],"Across":[109],"representative":[110],"dense":[111,176],"mixture-of-experts":[113],"workloads":[114],"both":[116],"prefill":[117],"decode,":[119],"we":[120],"show":[121,160],"latency,":[124],"GPU":[125],"inactivity,":[126],"ratios":[129],"alone":[130],"obscure":[132],"the":[133,150,234,238],"dominant":[134],"optimization":[135,142,230],"target.":[136],"instead":[138],"distinguishes":[139],"cases":[140,148],"where":[141,149],"reduce":[144],"software-stack":[145],"from":[147,154],"primary":[151],"win":[152],"comes":[153],"reducing":[155],"device-side":[156,239],"work.":[157],"further":[159],"MoE":[162],"models":[163],"dispatch":[164],"$\\mathbf{8":[165],"-":[166],"1":[167],"1}":[168],"\\times$":[169],"more":[170],"kernels":[171],"per":[172],"output":[173],"token":[174],"than":[175],"models,":[177],"such":[181],"host-bound":[182],"workloads,":[183],"CPU":[184,194],"singlethread":[185],"performance":[186],"first-order":[189],"parameter:":[190],"faster":[192],"host":[193],"reduces":[195],"$\\mathbf{1":[199],"0}-\\mathbf{2":[200],"9}":[201],"\\boldsymbol{\\%}$":[202],"improves":[204],"end-toend":[205],"latency":[206],"up":[208],"$14":[210],"\\%$,":[211],"even":[212],"when":[213],"paired":[214],"with":[215],"slowerclocked":[217],"GPU.":[218],"These":[219],"results":[220],"position":[221],"diagnostic":[225],"tool":[226],"assessing":[228],"whether":[229],"effort":[231],"target":[233],"software":[235],"stack":[236],"workload":[240],"execution.":[241]},"counts_by_year":[],"updated_date":"2026-06-11T09:08:48.828518","created_date":"2026-05-27T00:00:00"}