{"id":"https://openalex.org/W7117997745","doi":"https://doi.org/10.1145/3784828.3784830","title":"Understanding LLM Checkpoint/Restore I/O Strategies and Patterns","display_name":"Understanding LLM Checkpoint/Restore I/O Strategies and Patterns","publication_year":2026,"publication_date":"2026-01-22","ids":{"openalex":"https://openalex.org/W7117997745","doi":"https://doi.org/10.1145/3784828.3784830"},"language":null,"primary_location":{"id":"doi:10.1145/3784828.3784830","is_oa":true,"landing_page_url":"https://doi.org/10.1145/3784828.3784830","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":"Proceedings of the Supercomputing Asia and International Conference on High Performance Computing in Asia Pacific Region Workshops","raw_type":"proceedings-article"},"type":"conference-paper","indexed_in":["arxiv","crossref"],"open_access":{"is_oa":true,"oa_status":"gold","oa_url":"https://doi.org/10.1145/3784828.3784830","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5068912035","display_name":"Mikaila J. Gossman","orcid":"https://orcid.org/0009-0002-1668-0250"},"institutions":[{"id":"https://openalex.org/I8078737","display_name":"Clemson University","ror":"https://ror.org/037s24f05","country_code":"US","type":"education","lineage":["https://openalex.org/I8078737"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Mikaila Gossman","raw_affiliation_strings":["Clemson University, Clemson, SC, USA"],"raw_orcid":"https://orcid.org/0009-0002-1668-0250","affiliations":[{"raw_affiliation_string":"Clemson University, Clemson, SC, USA","institution_ids":["https://openalex.org/I8078737"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5049218913","display_name":"Avinash Maurya","orcid":"https://orcid.org/0000-0002-8200-0148"},"institutions":[{"id":"https://openalex.org/I1282105669","display_name":"Argonne National Laboratory","ror":"https://ror.org/05gvnxz63","country_code":"US","type":"facility","lineage":["https://openalex.org/I1282105669","https://openalex.org/I1330989302","https://openalex.org/I39565521","https://openalex.org/I40347166"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Avinash Maurya","raw_affiliation_strings":["Argonne National Lab, Lemont, IL, USA"],"raw_orcid":"https://orcid.org/0000-0002-8200-0148","affiliations":[{"raw_affiliation_string":"Argonne National Lab, Lemont, IL, USA","institution_ids":["https://openalex.org/I1282105669"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5121815467","display_name":"Bogdan Nicolae","orcid":null},"institutions":[{"id":"https://openalex.org/I1282105669","display_name":"Argonne National Laboratory","ror":"https://ror.org/05gvnxz63","country_code":"US","type":"facility","lineage":["https://openalex.org/I1282105669","https://openalex.org/I1330989302","https://openalex.org/I39565521","https://openalex.org/I40347166"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Bogdan Nicolae","raw_affiliation_strings":["Argonne National Lab, Lemont, IL, USA"],"raw_orcid":"https://orcid.org/0000-0002-0661-7509","affiliations":[{"raw_affiliation_string":"Argonne National Lab, Lemont, IL, USA","institution_ids":["https://openalex.org/I1282105669"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5003772863","display_name":"Jon C. Calhoun","orcid":"https://orcid.org/0000-0001-7191-4422"},"institutions":[{"id":"https://openalex.org/I8078737","display_name":"Clemson University","ror":"https://ror.org/037s24f05","country_code":"US","type":"education","lineage":["https://openalex.org/I8078737"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Jon Calhoun","raw_affiliation_strings":["Clemson University, Clemson, USA"],"raw_orcid":"https://orcid.org/0000-0001-7191-4422","affiliations":[{"raw_affiliation_string":"Clemson University, Clemson, USA","institution_ids":["https://openalex.org/I8078737"]}]}],"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":0.03793068,"is_in_top_1_percent":false,"is_in_top_10_percent":false},"cited_by_percentile_year":null,"biblio":{"volume":null,"issue":null,"first_page":"263","last_page":"273"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T11181","display_name":"Advanced Data Storage Technologies","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/T11181","display_name":"Advanced Data Storage Technologies","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/T10772","display_name":"Distributed systems and fault tolerance","score":0.0052999998442828655,"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/T10054","display_name":"Parallel Computing and Optimization Techniques","score":0.0024999999441206455,"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/throughput","display_name":"Throughput","score":0.6190000176429749},{"id":"https://openalex.org/keywords/asynchronous-communication","display_name":"Asynchronous communication","score":0.5950999855995178},{"id":"https://openalex.org/keywords/metadata","display_name":"Metadata","score":0.5806999802589417},{"id":"https://openalex.org/keywords/overhead","display_name":"Overhead (engineering)","score":0.5662000179290771},{"id":"https://openalex.org/keywords/traverse","display_name":"Traverse","score":0.4584999978542328},{"id":"https://openalex.org/keywords/data-access","display_name":"Data access","score":0.41359999775886536},{"id":"https://openalex.org/keywords/computer-data-storage","display_name":"Computer data storage","score":0.39800000190734863},{"id":"https://openalex.org/keywords/key","display_name":"Key (lock)","score":0.3935000002384186},{"id":"https://openalex.org/keywords/distributed-data-store","display_name":"Distributed data store","score":0.3896999955177307}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.8345000147819519},{"id":"https://openalex.org/C157764524","wikidata":"https://www.wikidata.org/wiki/Q1383412","display_name":"Throughput","level":3,"score":0.6190000176429749},{"id":"https://openalex.org/C151319957","wikidata":"https://www.wikidata.org/wiki/Q752739","display_name":"Asynchronous communication","level":2,"score":0.5950999855995178},{"id":"https://openalex.org/C93518851","wikidata":"https://www.wikidata.org/wiki/Q180160","display_name":"Metadata","level":2,"score":0.5806999802589417},{"id":"https://openalex.org/C2779960059","wikidata":"https://www.wikidata.org/wiki/Q7113681","display_name":"Overhead (engineering)","level":2,"score":0.5662000179290771},{"id":"https://openalex.org/C120314980","wikidata":"https://www.wikidata.org/wiki/Q180634","display_name":"Distributed computing","level":1,"score":0.5309000015258789},{"id":"https://openalex.org/C176809094","wikidata":"https://www.wikidata.org/wiki/Q15401496","display_name":"Traverse","level":2,"score":0.4584999978542328},{"id":"https://openalex.org/C47487241","wikidata":"https://www.wikidata.org/wiki/Q5227230","display_name":"Data access","level":2,"score":0.41359999775886536},{"id":"https://openalex.org/C173608175","wikidata":"https://www.wikidata.org/wiki/Q232661","display_name":"Parallel computing","level":1,"score":0.4025999903678894},{"id":"https://openalex.org/C194739806","wikidata":"https://www.wikidata.org/wiki/Q66221","display_name":"Computer data storage","level":2,"score":0.39800000190734863},{"id":"https://openalex.org/C26517878","wikidata":"https://www.wikidata.org/wiki/Q228039","display_name":"Key (lock)","level":2,"score":0.3935000002384186},{"id":"https://openalex.org/C24885549","wikidata":"https://www.wikidata.org/wiki/Q339678","display_name":"Distributed data store","level":2,"score":0.3896999955177307},{"id":"https://openalex.org/C2780940931","wikidata":"https://www.wikidata.org/wiki/Q174989","display_name":"File system","level":2,"score":0.3869999945163727},{"id":"https://openalex.org/C197115733","wikidata":"https://www.wikidata.org/wiki/Q1003136","display_name":"Forcing (mathematics)","level":2,"score":0.36800000071525574},{"id":"https://openalex.org/C188045654","wikidata":"https://www.wikidata.org/wiki/Q17148339","display_name":"Memory bandwidth","level":2,"score":0.3619999885559082},{"id":"https://openalex.org/C2776257435","wikidata":"https://www.wikidata.org/wiki/Q1576430","display_name":"Bandwidth (computing)","level":2,"score":0.3610999882221222},{"id":"https://openalex.org/C2781172179","wikidata":"https://www.wikidata.org/wiki/Q853109","display_name":"Parallelism (grammar)","level":2,"score":0.35910001397132874},{"id":"https://openalex.org/C111919701","wikidata":"https://www.wikidata.org/wiki/Q9135","display_name":"Operating system","level":1,"score":0.3154999911785126},{"id":"https://openalex.org/C2781057727","wikidata":"https://www.wikidata.org/wiki/Q9941","display_name":"Parallel I/O","level":2,"score":0.29829999804496765},{"id":"https://openalex.org/C2778755073","wikidata":"https://www.wikidata.org/wiki/Q10858537","display_name":"Scale (ratio)","level":2,"score":0.2808000147342682},{"id":"https://openalex.org/C31258907","wikidata":"https://www.wikidata.org/wiki/Q1301371","display_name":"Computer network","level":1,"score":0.27619999647140503},{"id":"https://openalex.org/C93996380","wikidata":"https://www.wikidata.org/wiki/Q44127","display_name":"Server","level":2,"score":0.2711000144481659},{"id":"https://openalex.org/C9395851","wikidata":"https://www.wikidata.org/wiki/Q177929","display_name":"Stack (abstract data type)","level":2,"score":0.2630999982357025},{"id":"https://openalex.org/C532756234","wikidata":"https://www.wikidata.org/wiki/Q14658","display_name":"POSIX","level":2,"score":0.25690001249313354},{"id":"https://openalex.org/C82687282","wikidata":"https://www.wikidata.org/wiki/Q66221","display_name":"Auxiliary memory","level":2,"score":0.2558000087738037},{"id":"https://openalex.org/C67646966","wikidata":"https://www.wikidata.org/wiki/Q237576","display_name":"Storage area network","level":3,"score":0.2515999972820282}],"mesh":[],"locations_count":2,"locations":[{"id":"doi:10.1145/3784828.3784830","is_oa":true,"landing_page_url":"https://doi.org/10.1145/3784828.3784830","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":"Proceedings of the Supercomputing Asia and International Conference on High Performance Computing in Asia Pacific Region Workshops","raw_type":"proceedings-article"},{"id":"pmh:oai:arXiv.org:2512.24511","is_oa":true,"landing_page_url":"http://arxiv.org/abs/2512.24511","pdf_url":"https://arxiv.org/pdf/2512.24511","source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"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"}],"best_oa_location":{"id":"doi:10.1145/3784828.3784830","is_oa":true,"landing_page_url":"https://doi.org/10.1145/3784828.3784830","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":"Proceedings of the Supercomputing Asia and International Conference on High Performance Computing in Asia Pacific Region Workshops","raw_type":"proceedings-article"},"sustainable_development_goals":[],"awards":[{"id":"https://openalex.org/G3357838806","display_name":"Collaborative Research: Elements: VLCC-States: Versioned Lineage-Driven Checkpointing of Composable States","funder_award_id":"2411386","funder_id":"https://openalex.org/F4320306076","funder_display_name":"National Science Foundation"},{"id":"https://openalex.org/G7140536155","display_name":"Collaborative Research: Elements: VLCC-States: Versioned Lineage-Driven Checkpointing of Composable States","funder_award_id":"2411387","funder_id":"https://openalex.org/F4320306076","funder_display_name":"National Science Foundation"}],"funders":[{"id":"https://openalex.org/F4320306076","display_name":"National Science Foundation","ror":"https://ror.org/021nxhr62"}],"has_content":{"grobid_xml":false,"pdf":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"As":[0,70],"large":[1,41,48],"language":[2],"models":[3,7],"(LLMs)":[4],"and":[5,18,26,55,74,84,111,182,184,211,217,237,258,274,286],"foundation":[6],"(FMs)":[8],"scale,":[9],"checkpoint-restore":[10],"has":[11],"become":[12],"critical":[13],"pattern":[14],"for":[15,180],"both":[16],"training":[17],"inferences.":[19],"This":[20],"is":[21,89],"a":[22,40,47,71,77,194],"difficult":[23],"pattern:":[24],"LLMs":[25],"FMs":[27],"are":[28],"distributed":[29],"at":[30],"scale":[31],"using":[32,203],"3D":[33],"parallelism":[34],"(tensor,":[35],"pipeline,":[36],"data),":[37],"resulting":[38],"in":[39,146,278],"number":[42,49],"of":[43,50,52,137,143,166,196,269],"processes,":[44],"each":[45],"holding":[46],"tensors":[51,103],"different":[53],"shapes":[54],"sizes,":[56],"which":[57],"need":[58,91,123],"to":[59,62,92,124,150,163,230,242,250,281],"be":[60,125],"captured":[61],"stable":[63],"storage":[64,96,110,139,285],"(e.g.":[65,131],"parallel":[66,117],"file":[67,118,271],"systems)":[68],"frequently.":[69],"consequence,":[72],"checkpoint":[73,183],"restore":[75,185],"becomes":[76],"big":[78],"data":[79],"problem,":[80],"exhibiting":[81],"volume,":[82],"variery":[83],"velocity.":[85],"A":[86],"key":[87],"challenge":[88],"the":[90,99,102,106,121,134,138,172,243,267],"traverse":[93],"an":[94],"entire":[95],"stack,":[97],"from":[98],"GPUs,":[100],"where":[101,120],"live,":[104],"through":[105],"host":[107],"memory,":[108],"local":[109],"finally":[112],"external":[113],"repositories":[114],"such":[115,159],"as":[116,160],"systems,":[119],"checkpoints":[122],"persisted.":[126],"Despite":[127],"efficient":[128],"I/O":[129,147,151,157,168,175,212,219,275,287],"techniques":[130],"asynchronous":[132],"flushing/prefetching),":[133],"memory":[135],"tiers":[136],"stack":[140],"exhibit":[141],"orders":[142],"magnitdue":[144],"difference":[145],"performance,":[148],"leading":[149],"bottlenecks,":[152],"especially":[153],"under":[154,215],"concurrency.":[155],"Kernel-accelerated":[156],"libraries":[158],"liburing":[161],"aim":[162],"address":[164],"some":[165],"these":[167],"bottlenecks":[169],"compared":[170,241],"with":[171],"aging":[173],"POSIX":[174],"interface.":[176],"However,":[177],"their":[178],"effectiveness":[179],"LLM":[181,245,279],"remains":[186],"largely":[187],"unexplored.":[188],"To":[189],"this":[190],"end,":[191],"we":[192,206],"develop":[193],"series":[195],"microbenchmarks":[197],"that":[198,223],"study":[199],"various":[200],"trade-off":[201],"when":[202],"liburing.":[204],"Specifically,":[205],"evaluate":[207],"how":[208],"aggregation,":[209],"alignment,":[210],"coalescing":[213,276],"interact":[214],"buffered":[216],"direct":[218],"modes.":[220],"Results":[221],"show":[222],"uncoalesced,":[224],"small-buffer":[225],"operations":[226],"halve":[227],"throughput":[228,255],"relative":[229],"synthetic":[231],"workloads,":[232],"while":[233],"aggregation":[234,273],"restores":[235],"bandwidth":[236],"reduces":[238],"metadata":[239],"overhead":[240],"state-of-the-art":[244],"checkpointing":[246,280],"engines,":[247],"achieving":[248],"up":[249],"3.9":[251],"\u00d7":[252,260],"higher":[253,261],"write":[254],"than":[256,262],"DataStates-LLM":[257],"7.6":[259],"TorchSnapshot.":[263],"These":[264],"findings":[265],"underscore":[266],"importance":[268],"adopting":[270],"system\u2013aware":[272],"strategies":[277],"fully":[282],"exploit":[283],"modern":[284],"backends.":[288]},"counts_by_year":[],"updated_date":"2026-08-05T07:39:15.569665","created_date":"2026-01-02T00:00:00"}