{"id":"https://openalex.org/W7128506609","doi":"https://doi.org/10.1109/tetc.2026.3661468","title":"MemAscend: System Memory Optimization for SSD-Offloaded LLM Fine-Tuning","display_name":"MemAscend: System Memory Optimization for SSD-Offloaded LLM Fine-Tuning","publication_year":2026,"publication_date":"2026-01-01","ids":{"openalex":"https://openalex.org/W7128506609","doi":"https://doi.org/10.1109/tetc.2026.3661468"},"language":null,"primary_location":{"id":"doi:10.1109/tetc.2026.3661468","is_oa":false,"landing_page_url":"https://doi.org/10.1109/tetc.2026.3661468","pdf_url":null,"source":{"id":"https://openalex.org/S2496326734","display_name":"IEEE Transactions on Emerging Topics in Computing","issn_l":"2168-6750","issn":["2168-6750","2376-4562"],"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 Emerging Topics in 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":"https://openalex.org/A5104171037","display_name":"Yong-Cheng Liaw","orcid":null},"institutions":[{"id":"https://openalex.org/I148366613","display_name":"National Yang Ming Chiao Tung University","ror":"https://ror.org/00se2k293","country_code":"TW","type":"education","lineage":["https://openalex.org/I148366613"]}],"countries":["TW"],"is_corresponding":true,"raw_author_name":"Yong-Cheng Liaw","raw_affiliation_strings":["Institute of Computer Science and Engineering, National Yang Ming Chiao Tung University, Hsinchu, Taiwan"],"raw_orcid":"https://orcid.org/0009-0007-9783-4640","affiliations":[{"raw_affiliation_string":"Institute of Computer Science and Engineering, National Yang Ming Chiao Tung University, Hsinchu, Taiwan","institution_ids":["https://openalex.org/I148366613"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5060246506","display_name":"Shuo-Han Chen","orcid":"https://orcid.org/0000-0002-1619-4335"},"institutions":[{"id":"https://openalex.org/I148366613","display_name":"National Yang Ming Chiao Tung University","ror":"https://ror.org/00se2k293","country_code":"TW","type":"education","lineage":["https://openalex.org/I148366613"]}],"countries":["TW"],"is_corresponding":false,"raw_author_name":"Shuo-Han Chen","raw_affiliation_strings":["Department of Computer Science and the Institute of Artificial Intelligence Innovation, National Yang Ming Chiao Tung University, Hsinchu, Taiwan"],"raw_orcid":"https://orcid.org/0000-0002-1619-4335","affiliations":[{"raw_affiliation_string":"Department of Computer Science and the Institute of Artificial Intelligence Innovation, National Yang Ming Chiao Tung University, Hsinchu, Taiwan","institution_ids":["https://openalex.org/I148366613"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":2,"corresponding_author_ids":["https://openalex.org/A5104171037"],"corresponding_institution_ids":["https://openalex.org/I148366613"],"apc_list":null,"apc_paid":null,"fwci":0.0,"has_fulltext":false,"cited_by_count":0,"citation_normalized_percentile":{"value":0.2454644,"is_in_top_1_percent":false,"is_in_top_10_percent":false},"cited_by_percentile_year":null,"biblio":{"volume":"14","issue":"1","first_page":"286","last_page":"302"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T11181","display_name":"Advanced Data Storage Technologies","score":0.7937999963760376,"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":0.7937999963760376,"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":0.0706000030040741,"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":0.048500001430511475,"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/scalability","display_name":"Scalability","score":0.6438999772071838},{"id":"https://openalex.org/keywords/memory-management","display_name":"Memory management","score":0.5418000221252441},{"id":"https://openalex.org/keywords/flat-memory-model","display_name":"Flat memory model","score":0.5163000226020813},{"id":"https://openalex.org/keywords/file-system","display_name":"File system","score":0.4819999933242798},{"id":"https://openalex.org/keywords/key","display_name":"Key (lock)","score":0.47119998931884766},{"id":"https://openalex.org/keywords/context","display_name":"Context (archaeology)","score":0.4699000120162964},{"id":"https://openalex.org/keywords/interleaved-memory","display_name":"Interleaved memory","score":0.4075999855995178},{"id":"https://openalex.org/keywords/non-volatile-memory","display_name":"Non-volatile memory","score":0.4027999937534332},{"id":"https://openalex.org/keywords/computing-with-memory","display_name":"Computing with Memory","score":0.398499995470047},{"id":"https://openalex.org/keywords/multi-core-processor","display_name":"Multi-core processor","score":0.3833000063896179}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.8690999746322632},{"id":"https://openalex.org/C48044578","wikidata":"https://www.wikidata.org/wiki/Q727490","display_name":"Scalability","level":2,"score":0.6438999772071838},{"id":"https://openalex.org/C176649486","wikidata":"https://www.wikidata.org/wiki/Q2308807","display_name":"Memory management","level":3,"score":0.5418000221252441},{"id":"https://openalex.org/C57863822","wikidata":"https://www.wikidata.org/wiki/Q905488","display_name":"Flat memory model","level":4,"score":0.5163000226020813},{"id":"https://openalex.org/C2780940931","wikidata":"https://www.wikidata.org/wiki/Q174989","display_name":"File system","level":2,"score":0.4819999933242798},{"id":"https://openalex.org/C26517878","wikidata":"https://www.wikidata.org/wiki/Q228039","display_name":"Key (lock)","level":2,"score":0.47119998931884766},{"id":"https://openalex.org/C149635348","wikidata":"https://www.wikidata.org/wiki/Q193040","display_name":"Embedded system","level":1,"score":0.47119998931884766},{"id":"https://openalex.org/C2779343474","wikidata":"https://www.wikidata.org/wiki/Q3109175","display_name":"Context (archaeology)","level":2,"score":0.4699000120162964},{"id":"https://openalex.org/C118524514","wikidata":"https://www.wikidata.org/wiki/Q173212","display_name":"Computer architecture","level":1,"score":0.41449999809265137},{"id":"https://openalex.org/C63511323","wikidata":"https://www.wikidata.org/wiki/Q908936","display_name":"Interleaved memory","level":4,"score":0.4075999855995178},{"id":"https://openalex.org/C177950962","wikidata":"https://www.wikidata.org/wiki/Q10997658","display_name":"Non-volatile memory","level":2,"score":0.4027999937534332},{"id":"https://openalex.org/C152890283","wikidata":"https://www.wikidata.org/wiki/Q4129922","display_name":"Computing with Memory","level":5,"score":0.398499995470047},{"id":"https://openalex.org/C78766204","wikidata":"https://www.wikidata.org/wiki/Q555032","display_name":"Multi-core processor","level":2,"score":0.3833000063896179},{"id":"https://openalex.org/C18131444","wikidata":"https://www.wikidata.org/wiki/Q163585","display_name":"Memory protection","level":5,"score":0.36250001192092896},{"id":"https://openalex.org/C31352089","wikidata":"https://www.wikidata.org/wiki/Q3750474","display_name":"Systems design","level":2,"score":0.3610999882221222},{"id":"https://openalex.org/C9390403","wikidata":"https://www.wikidata.org/wiki/Q3966","display_name":"Computer hardware","level":1,"score":0.35409998893737793},{"id":"https://openalex.org/C192209626","wikidata":"https://www.wikidata.org/wiki/Q190909","display_name":"Focus (optics)","level":2,"score":0.3458000123500824},{"id":"https://openalex.org/C188198153","wikidata":"https://www.wikidata.org/wiki/Q1613840","display_name":"Limiting","level":2,"score":0.34209999442100525},{"id":"https://openalex.org/C118702147","wikidata":"https://www.wikidata.org/wiki/Q189396","display_name":"Dynamic random-access memory","level":3,"score":0.33629998564720154},{"id":"https://openalex.org/C171675096","wikidata":"https://www.wikidata.org/wiki/Q1143380","display_name":"Extended memory","level":4,"score":0.32820001244544983},{"id":"https://openalex.org/C74426580","wikidata":"https://www.wikidata.org/wiki/Q719484","display_name":"Memory map","level":3,"score":0.3246000111103058},{"id":"https://openalex.org/C120314980","wikidata":"https://www.wikidata.org/wiki/Q180634","display_name":"Distributed computing","level":1,"score":0.3181999921798706},{"id":"https://openalex.org/C173608175","wikidata":"https://www.wikidata.org/wiki/Q232661","display_name":"Parallel computing","level":1,"score":0.31790000200271606},{"id":"https://openalex.org/C118021083","wikidata":"https://www.wikidata.org/wiki/Q610398","display_name":"System on a chip","level":2,"score":0.3154999911785126},{"id":"https://openalex.org/C2779602883","wikidata":"https://www.wikidata.org/wiki/Q15544750","display_name":"Memory architecture","level":2,"score":0.3133000135421753},{"id":"https://openalex.org/C51290061","wikidata":"https://www.wikidata.org/wiki/Q1936765","display_name":"Uniform memory access","level":4,"score":0.311599999666214},{"id":"https://openalex.org/C2780165032","wikidata":"https://www.wikidata.org/wiki/Q16869822","display_name":"Energy consumption","level":2,"score":0.30959999561309814},{"id":"https://openalex.org/C82687282","wikidata":"https://www.wikidata.org/wiki/Q66221","display_name":"Auxiliary memory","level":2,"score":0.29820001125335693},{"id":"https://openalex.org/C91481028","wikidata":"https://www.wikidata.org/wiki/Q1054686","display_name":"Distributed memory","level":3,"score":0.29350000619888306},{"id":"https://openalex.org/C194739806","wikidata":"https://www.wikidata.org/wiki/Q66221","display_name":"Computer data storage","level":2,"score":0.2865999937057495},{"id":"https://openalex.org/C133875982","wikidata":"https://www.wikidata.org/wiki/Q764810","display_name":"Shared memory","level":2,"score":0.2702000141143799},{"id":"https://openalex.org/C153247305","wikidata":"https://www.wikidata.org/wiki/Q835713","display_name":"Memory address","level":3,"score":0.2685999870300293},{"id":"https://openalex.org/C2776760102","wikidata":"https://www.wikidata.org/wiki/Q5139990","display_name":"Code (set theory)","level":3,"score":0.2662999927997589},{"id":"https://openalex.org/C39528615","wikidata":"https://www.wikidata.org/wiki/Q1229610","display_name":"Distributed shared memory","level":5,"score":0.25940001010894775},{"id":"https://openalex.org/C111919701","wikidata":"https://www.wikidata.org/wiki/Q9135","display_name":"Operating system","level":1,"score":0.2581999897956848},{"id":"https://openalex.org/C2781357197","wikidata":"https://www.wikidata.org/wiki/Q5757597","display_name":"High memory","level":2,"score":0.2578999996185303},{"id":"https://openalex.org/C12186640","wikidata":"https://www.wikidata.org/wiki/Q6815743","display_name":"Memory model","level":3,"score":0.25760000944137573},{"id":"https://openalex.org/C98986596","wikidata":"https://www.wikidata.org/wiki/Q1143031","display_name":"Semiconductor memory","level":2,"score":0.2574000060558319},{"id":"https://openalex.org/C41036726","wikidata":"https://www.wikidata.org/wiki/Q844824","display_name":"Physical address","level":3,"score":0.2572999894618988}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.1109/tetc.2026.3661468","is_oa":false,"landing_page_url":"https://doi.org/10.1109/tetc.2026.3661468","pdf_url":null,"source":{"id":"https://openalex.org/S2496326734","display_name":"IEEE Transactions on Emerging Topics in Computing","issn_l":"2168-6750","issn":["2168-6750","2376-4562"],"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 Emerging Topics in Computing","raw_type":"journal-article"}],"best_oa_location":null,"sustainable_development_goals":[],"awards":[],"funders":[],"has_content":{"pdf":false,"grobid_xml":false},"content_urls":null,"referenced_works_count":5,"referenced_works":["https://openalex.org/W3129831491","https://openalex.org/W3205803342","https://openalex.org/W4289646389","https://openalex.org/W4393407018","https://openalex.org/W7133227460"],"related_works":[],"abstract_inverted_index":{"Owing":[0],"to":[1,69,133],"the":[2,71,141,211],"huge":[3],"success":[4],"of":[5,202],"generative":[6],"artificial":[7],"intelligence":[8],"(AI),":[9],"large":[10],"language":[11],"models":[12,34],"(LLMs)":[13],"have":[14],"emerged":[15,64],"as":[16,23,65],"a":[17,66,98,136,151,170],"core":[18],"subclass,":[19],"powering":[20],"applications":[21],"such":[22],"question":[24],"answering,":[25],"text":[26],"generation,":[27],"and":[28,52,83,116,124,162,181,216],"code":[29],"completion.":[30],"While":[31],"fine-tuning":[32,215],"these":[33],"on":[35,153,224],"domain-specific":[36],"data":[37],"can":[38],"yield":[39],"significant":[40],"performance":[41,95],"gains,":[42],"it":[43],"also":[44],"poses":[45],"daunting":[46],"computational":[47],"challenges,":[48],"especially":[49],"for":[50,214,220],"researchers":[51],"organizations":[53],"with":[54,150,205],"limited":[55],"hardware":[56,188,212],"resources.":[57],"Although":[58],"SSD":[59,207],"offloading":[60,208],"(i.e.,":[61,80,86],"ZeRO-Infinity)":[62],"has":[63],"viable":[67],"strategy":[68],"overcome":[70],"GPU":[72],"memory":[73,79,105,113,144,166,173],"barrier":[74,213],"via":[75],"leveraging":[76],"both":[77],"system":[78,104,112,118,143,165,172],"CPU":[81],"DRAM)":[82],"storage":[84],"space":[85],"solid-state":[87],"devices,":[88],"SSDs),":[89],"its":[90],"design":[91],"primarily":[92],"targets":[93],"model-centric":[94],"issues.":[96],"As":[97],"result,":[99],"key":[100],"system-level":[101],"issues,":[102],"including":[103],"fragmentation,":[106,161],"inefficient":[107],"pinned":[108],"buffer":[109],"allocation,":[110,159],"peak":[111,164,196],"usage":[114],"spikes,":[115],"file":[117],"overhead,":[119],"remain":[120],"unaddressed,":[121],"limiting":[122],"scalability":[123],"inflating":[125],"costs.":[126],"Such":[127],"an":[128,200],"observation":[129],"motivates":[130],"this":[131],"paper":[132],"introduce":[134],"MemAscend,":[135],"framework":[137],"that":[138],"systematically":[139],"tackles":[140],"underexplored":[142],"bottlenecks":[145],"in":[146],"SSD-offloaded":[147],"LLM":[148,192],"training,":[149],"focus":[152],"resource-constrained":[154],"environments.":[155],"By":[156],"streamlining":[157],"pinned-memory":[158],"eradicating":[160],"mitigating":[163],"usage,":[167],"MemAscend":[168,194],"reclaims":[169],"substantial":[171],"budget,":[174],"enabling":[175],"larger":[176],"models,":[177],"longer":[178],"context":[179],"windows,":[180],"higher":[182],"batch":[183],"sizes":[184],"without":[185],"exceeding":[186],"modest":[187],"limits.":[189],"Across":[190],"diverse":[191],"benchmarks,":[193],"reduces":[195],"system-memory":[197],"consumption":[198],"by":[199],"average":[201],"55.7%":[203],"compared":[204],"state-of-the-art":[206],"techniques,":[209],"lowering":[210],"unlocking":[217],"new":[218],"possibilities":[219],"cost-effective":[221],"large-scale":[222],"training":[223],"limited-resource":[225],"machines.":[226]},"counts_by_year":[],"updated_date":"2026-03-14T06:41:57.775601","created_date":"2026-02-06T00:00:00"}