{"id":"https://openalex.org/W7116919423","doi":"https://doi.org/10.48550/arxiv.2512.17570","title":"GreedySnake: Accelerating SSD-Offloaded LLM Training with Efficient Scheduling and Optimizer Step Overlapping","display_name":"GreedySnake: Accelerating SSD-Offloaded LLM Training with Efficient Scheduling and Optimizer Step Overlapping","publication_year":2025,"publication_date":"2025-12-19","ids":{"openalex":"https://openalex.org/W7116919423","doi":"https://doi.org/10.48550/arxiv.2512.17570"},"language":null,"primary_location":{"id":"doi:10.48550/arxiv.2512.17570","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2512.17570","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":null,"issn":null,"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":null,"is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"Preprint"},"type":"preprint","indexed_in":["datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://doi.org/10.48550/arxiv.2512.17570","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":null,"display_name":"Yin, Yishu","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Yin, Yishu","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":null,"display_name":"Qian, Xuehai","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Qian, Xuehai","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]}],"institutions":[],"countries_distinct_count":0,"institutions_distinct_count":0,"corresponding_author_ids":[],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":null,"has_fulltext":false,"cited_by_count":0,"citation_normalized_percentile":null,"cited_by_percentile_year":null,"biblio":{"volume":null,"issue":null,"first_page":null,"last_page":null},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T10054","display_name":"Parallel Computing and Optimization Techniques","score":0.4334999918937683,"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":0.4334999918937683,"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/T11181","display_name":"Advanced Data Storage Technologies","score":0.3612000048160553,"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/T10101","display_name":"Cloud Computing and Resource Management","score":0.041600000113248825,"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/training","display_name":"Training (meteorology)","score":0.7357000112533569},{"id":"https://openalex.org/keywords/scheduling","display_name":"Scheduling (production processes)","score":0.6985999941825867},{"id":"https://openalex.org/keywords/throughput","display_name":"Throughput","score":0.477400004863739},{"id":"https://openalex.org/keywords/ideal","display_name":"Ideal (ethics)","score":0.398499995470047},{"id":"https://openalex.org/keywords/layer","display_name":"Layer (electronics)","score":0.3874000012874603},{"id":"https://openalex.org/keywords/training-set","display_name":"Training set","score":0.3476000130176544}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.7806000113487244},{"id":"https://openalex.org/C2777211547","wikidata":"https://www.wikidata.org/wiki/Q17141490","display_name":"Training (meteorology)","level":2,"score":0.7357000112533569},{"id":"https://openalex.org/C206729178","wikidata":"https://www.wikidata.org/wiki/Q2271896","display_name":"Scheduling (production processes)","level":2,"score":0.6985999941825867},{"id":"https://openalex.org/C157764524","wikidata":"https://www.wikidata.org/wiki/Q1383412","display_name":"Throughput","level":3,"score":0.477400004863739},{"id":"https://openalex.org/C173608175","wikidata":"https://www.wikidata.org/wiki/Q232661","display_name":"Parallel computing","level":1,"score":0.4442000091075897},{"id":"https://openalex.org/C2776639384","wikidata":"https://www.wikidata.org/wiki/Q840396","display_name":"Ideal (ethics)","level":2,"score":0.398499995470047},{"id":"https://openalex.org/C2779227376","wikidata":"https://www.wikidata.org/wiki/Q6505497","display_name":"Layer (electronics)","level":2,"score":0.3874000012874603},{"id":"https://openalex.org/C79403827","wikidata":"https://www.wikidata.org/wiki/Q3988","display_name":"Real-time computing","level":1,"score":0.36980000138282776},{"id":"https://openalex.org/C120314980","wikidata":"https://www.wikidata.org/wiki/Q180634","display_name":"Distributed computing","level":1,"score":0.34769999980926514},{"id":"https://openalex.org/C51632099","wikidata":"https://www.wikidata.org/wiki/Q3985153","display_name":"Training set","level":2,"score":0.3476000130176544},{"id":"https://openalex.org/C2984822820","wikidata":"https://www.wikidata.org/wiki/Q1123036","display_name":"Processor scheduling","level":3,"score":0.29899999499320984},{"id":"https://openalex.org/C118524514","wikidata":"https://www.wikidata.org/wiki/Q173212","display_name":"Computer architecture","level":1,"score":0.2987000048160553},{"id":"https://openalex.org/C149635348","wikidata":"https://www.wikidata.org/wiki/Q193040","display_name":"Embedded system","level":1,"score":0.2930999994277954},{"id":"https://openalex.org/C55416958","wikidata":"https://www.wikidata.org/wiki/Q6206757","display_name":"Job shop scheduling","level":3,"score":0.28060001134872437},{"id":"https://openalex.org/C113775141","wikidata":"https://www.wikidata.org/wiki/Q428691","display_name":"Computer engineering","level":1,"score":0.27619999647140503},{"id":"https://openalex.org/C2776857766","wikidata":"https://www.wikidata.org/wiki/Q7832987","display_name":"Training system","level":2,"score":0.26409998536109924},{"id":"https://openalex.org/C2989134064","wikidata":"https://www.wikidata.org/wiki/Q288510","display_name":"Execution time","level":2,"score":0.2554999887943268},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.25119999051094055}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2512.17570","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2512.17570","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":null,"issn":null,"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":null,"is_accepted":false,"is_published":null,"raw_source_name":null,"raw_type":"Preprint"}],"best_oa_location":{"id":"doi:10.48550/arxiv.2512.17570","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2512.17570","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":null,"issn":null,"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":null,"is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"Preprint"},"sustainable_development_goals":[],"awards":[],"funders":[],"has_content":{"pdf":false,"grobid_xml":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"SSD-offloaded":[0,25],"training":[1,11,26,59,110],"offers":[2],"a":[3,23,37],"practical":[4],"and":[5,119,126],"promising":[6],"approach":[7],"to":[8,41,45,70],"making":[9],"LLM":[10],"cost-effective.":[12],"Building":[13],"on":[14,102,116,121,128],"gradient":[15],"accumulation":[16],"with":[17,61,92],"micro-batches,":[18],"this":[19],"paper":[20],"introduces":[21],"GreedySnake,":[22],"new":[24],"system":[27,67],"that":[28,48,106],"employs":[29],"vertical":[30],"scheduling,":[31],"which":[32],"executes":[33],"all":[34],"microbatches":[35],"of":[36,88,96],"layer":[38],"before":[39],"proceeding":[40],"the":[42,66,71,76,82,89,93,97],"next.":[43],"Compared":[44],"existing":[46],"systems":[47],"use":[49],"horizontal":[50],"scheduling":[51],"(i.e.,":[52],"executing":[53],"micro-batches":[54],"sequentially),":[55],"GreedySnake":[56,85,107],"achieves":[57,108],"higher":[58],"throughput":[60,111],"smaller":[62],"batch":[63],"sizes,":[64],"bringing":[65],"much":[68],"closer":[69],"ideal":[72],"scenario":[73],"predicted":[74],"by":[75],"roofline":[77],"model.":[78],"To":[79],"further":[80],"mitigate":[81],"I/O":[83],"bottleneck,":[84],"overlaps":[86],"part":[87],"optimization":[90],"step":[91],"forward":[94],"pass":[95],"next":[98],"iteration.":[99],"Experimental":[100],"results":[101],"A100":[103],"GPUs":[104,123],"show":[105],"saturated":[109],"improvements":[112],"over":[113],"ZeRO-Infinity:":[114],"1.96x":[115],"1":[117,129],"GPU":[118,130],"1.93x":[120],"4":[122],"for":[124,131],"GPT-65B,":[125],"2.53x":[127],"GPT-175B.":[132]},"counts_by_year":[],"updated_date":"2026-07-01T06:00:48.157686","created_date":"2025-12-23T00:00:00"}