{"id":"https://openalex.org/W4386260508","doi":"https://doi.org/10.1109/infocom53939.2023.10228922","title":"Libra: Contention-Aware GPU Thread Allocation for Data Parallel Training in High Speed Networks","display_name":"Libra: Contention-Aware GPU Thread Allocation for Data Parallel Training in High Speed Networks","publication_year":2023,"publication_date":"2023-05-17","ids":{"openalex":"https://openalex.org/W4386260508","doi":"https://doi.org/10.1109/infocom53939.2023.10228922"},"language":"en","primary_location":{"id":"doi:10.1109/infocom53939.2023.10228922","is_oa":false,"landing_page_url":"http://dx.doi.org/10.1109/infocom53939.2023.10228922","pdf_url":null,"source":null,"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"IEEE INFOCOM 2023 - IEEE Conference on Computer Communications","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/A5052085622","display_name":"Yunzhuo Liu","orcid":"https://orcid.org/0000-0003-0156-2454"},"institutions":[{"id":"https://openalex.org/I183067930","display_name":"Shanghai Jiao Tong University","ror":"https://ror.org/0220qvk04","country_code":"CN","type":"education","lineage":["https://openalex.org/I183067930"]}],"countries":["CN"],"is_corresponding":true,"raw_author_name":"Yunzhuo Liu","raw_affiliation_strings":["Shanghai Jiao Tong University"],"affiliations":[{"raw_affiliation_string":"Shanghai Jiao Tong University","institution_ids":["https://openalex.org/I183067930"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5100669526","display_name":"Bo Jiang","orcid":"https://orcid.org/0000-0001-6711-4342"},"institutions":[{"id":"https://openalex.org/I183067930","display_name":"Shanghai Jiao Tong University","ror":"https://ror.org/0220qvk04","country_code":"CN","type":"education","lineage":["https://openalex.org/I183067930"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Bo Jiang","raw_affiliation_strings":["Shanghai Jiao Tong University"],"affiliations":[{"raw_affiliation_string":"Shanghai Jiao Tong University","institution_ids":["https://openalex.org/I183067930"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5075325244","display_name":"Shizhen Zhao","orcid":"https://orcid.org/0000-0001-8395-5109"},"institutions":[{"id":"https://openalex.org/I183067930","display_name":"Shanghai Jiao Tong University","ror":"https://ror.org/0220qvk04","country_code":"CN","type":"education","lineage":["https://openalex.org/I183067930"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Shizhen Zhao","raw_affiliation_strings":["Shanghai Jiao Tong University"],"affiliations":[{"raw_affiliation_string":"Shanghai Jiao Tong University","institution_ids":["https://openalex.org/I183067930"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5101944426","display_name":"Tao Lin","orcid":"https://orcid.org/0000-0003-1170-636X"},"institutions":[{"id":"https://openalex.org/I75689368","display_name":"Communication University of China","ror":"https://ror.org/04facbs33","country_code":"CN","type":"education","lineage":["https://openalex.org/I75689368"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Tao Lin","raw_affiliation_strings":["Communication University of China"],"affiliations":[{"raw_affiliation_string":"Communication University of China","institution_ids":["https://openalex.org/I75689368"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5034483183","display_name":"Xinbing Wang","orcid":"https://orcid.org/0000-0002-0357-8356"},"institutions":[{"id":"https://openalex.org/I183067930","display_name":"Shanghai Jiao Tong University","ror":"https://ror.org/0220qvk04","country_code":"CN","type":"education","lineage":["https://openalex.org/I183067930"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Xinbing Wang","raw_affiliation_strings":["Shanghai Jiao Tong University"],"affiliations":[{"raw_affiliation_string":"Shanghai Jiao Tong University","institution_ids":["https://openalex.org/I183067930"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5023919188","display_name":"Chenghu Zhou","orcid":"https://orcid.org/0000-0003-3331-2302"},"institutions":[{"id":"https://openalex.org/I19820366","display_name":"Chinese Academy of Sciences","ror":"https://ror.org/034t30j35","country_code":"CN","type":"funder","lineage":["https://openalex.org/I19820366"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Chenghu Zhou","raw_affiliation_strings":["Chinese Academy of Sciences"],"affiliations":[{"raw_affiliation_string":"Chinese Academy of Sciences","institution_ids":["https://openalex.org/I19820366"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":6,"corresponding_author_ids":["https://openalex.org/A5052085622"],"corresponding_institution_ids":["https://openalex.org/I183067930"],"apc_list":null,"apc_paid":null,"fwci":0.369,"has_fulltext":false,"cited_by_count":3,"citation_normalized_percentile":{"value":0.59847832,"is_in_top_1_percent":false,"is_in_top_10_percent":false},"cited_by_percentile_year":{"min":89,"max":95},"biblio":{"volume":"31","issue":null,"first_page":"1","last_page":"10"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T10036","display_name":"Advanced Neural Network Applications","score":0.9988999962806702,"subfield":{"id":"https://openalex.org/subfields/1707","display_name":"Computer Vision and Pattern Recognition"},"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/T10036","display_name":"Advanced Neural Network Applications","score":0.9988999962806702,"subfield":{"id":"https://openalex.org/subfields/1707","display_name":"Computer Vision and Pattern Recognition"},"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":0.9789000153541565,"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/T12702","display_name":"Brain Tumor Detection and Classification","score":0.9753000140190125,"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"}}],"keywords":[{"id":"https://openalex.org/keywords/computer-science","display_name":"Computer science","score":0.8642438650131226},{"id":"https://openalex.org/keywords/thread","display_name":"Thread (computing)","score":0.73917555809021},{"id":"https://openalex.org/keywords/parallel-computing","display_name":"Parallel computing","score":0.7079894542694092},{"id":"https://openalex.org/keywords/operating-system","display_name":"Operating system","score":0.24673500657081604}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.8642438650131226},{"id":"https://openalex.org/C138101251","wikidata":"https://www.wikidata.org/wiki/Q213092","display_name":"Thread (computing)","level":2,"score":0.73917555809021},{"id":"https://openalex.org/C173608175","wikidata":"https://www.wikidata.org/wiki/Q232661","display_name":"Parallel computing","level":1,"score":0.7079894542694092},{"id":"https://openalex.org/C111919701","wikidata":"https://www.wikidata.org/wiki/Q9135","display_name":"Operating system","level":1,"score":0.24673500657081604}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.1109/infocom53939.2023.10228922","is_oa":false,"landing_page_url":"http://dx.doi.org/10.1109/infocom53939.2023.10228922","pdf_url":null,"source":null,"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"IEEE INFOCOM 2023 - IEEE Conference on Computer Communications","raw_type":"proceedings-article"}],"best_oa_location":null,"sustainable_development_goals":[],"awards":[],"funders":[{"id":"https://openalex.org/F4320321001","display_name":"National Natural Science Foundation of China","ror":"https://ror.org/01h0zpd94"},{"id":"https://openalex.org/F4320329139","display_name":"Communication University of China","ror":"https://ror.org/04facbs33"}],"has_content":{"grobid_xml":false,"pdf":false},"content_urls":null,"referenced_works_count":41,"referenced_works":["https://openalex.org/W1566289585","https://openalex.org/W1686810756","https://openalex.org/W2057332538","https://openalex.org/W2108598243","https://openalex.org/W2131613942","https://openalex.org/W2194775991","https://openalex.org/W2339765813","https://openalex.org/W2531409750","https://openalex.org/W2745269232","https://openalex.org/W2766140019","https://openalex.org/W2786171709","https://openalex.org/W2891800530","https://openalex.org/W2896457183","https://openalex.org/W2906007643","https://openalex.org/W2946900926","https://openalex.org/W2962758826","https://openalex.org/W2975712713","https://openalex.org/W3047357290","https://openalex.org/W3047537431","https://openalex.org/W3086105743","https://openalex.org/W3107997909","https://openalex.org/W3124352525","https://openalex.org/W3133730715","https://openalex.org/W3177554119","https://openalex.org/W3188065709","https://openalex.org/W3193250980","https://openalex.org/W4224214704","https://openalex.org/W4283217358","https://openalex.org/W4288357791","https://openalex.org/W4289401659","https://openalex.org/W4301239768","https://openalex.org/W6637373629","https://openalex.org/W6739693220","https://openalex.org/W6745723224","https://openalex.org/W6748645090","https://openalex.org/W6754799804","https://openalex.org/W6754818105","https://openalex.org/W6755207826","https://openalex.org/W6762211661","https://openalex.org/W6763138067","https://openalex.org/W6810097121"],"related_works":["https://openalex.org/W4391375266","https://openalex.org/W2748952813","https://openalex.org/W2390279801","https://openalex.org/W2358668433","https://openalex.org/W4396701345","https://openalex.org/W2376932109","https://openalex.org/W2001405890","https://openalex.org/W4396696052","https://openalex.org/W2023832055","https://openalex.org/W2082485924"],"abstract_inverted_index":{"Overlapping":[0],"gradient":[1],"communication":[2,12,30,68,99,109],"with":[3,132,147,167,172,184],"backward":[4,115,162],"computation":[5,27,97],"is":[6],"a":[7,119,185],"popular":[8],"technique":[9],"to":[10,48,57,71,89,112,128],"reduce":[11,113],"cost":[13],"in":[14,31],"the":[15,23,35,60,63,91,95,108,181,191],"widely":[16],"adopted":[17],"data":[18],"parallel":[19],"S-SGD":[20],"training.":[21],"However,":[22],"resource":[24],"contention":[25,64,75,92],"between":[26],"and":[28,80,98,125,139],"All-Reduce":[29],"GPU-based":[32],"training":[33,151,186],"reduces":[34,161],"benefits":[36],"of":[37,188],"overlap.":[38],"With":[39],"GPU":[40,53,78,145],"cluster":[41,146,198],"network":[42,150],"evolving":[43],"from":[44],"low":[45],"bandwidth":[46],"TCP":[47],"high":[49],"speed":[50],"networks,":[51],"more":[52,65],"resources":[54],"are":[55],"required":[56],"efficiently":[58],"utilize":[59],"bandwidth,":[61],"making":[62],"noticeable.":[66],"Existing":[67],"libraries":[69],"fail":[70],"account":[72],"for":[73],"such":[74],"when":[76],"allocating":[77],"threads":[79],"have":[81],"suboptimal":[82],"performance.":[83],"In":[84],"this":[85],"paper,":[86],"we":[87,195],"propose":[88],"mitigate":[90],"by":[93,164,169],"balancing":[94],"overlapped":[96],"time.":[100,116],"We":[101,117,135],"formulate":[102],"an":[103,143],"optimization":[104],"problem":[105],"that":[106,158,177],"decides":[107],"thread":[110,130],"allocation":[111,131],"overall":[114],"develop":[118],"dynamic":[120],"programming":[121],"based":[122],"near-optimal":[123],"solution":[124],"extend":[126],"it":[127],"co-optimize":[129],"tensor":[133],"fusion.":[134],"conduct":[136],"simulated":[137],"study":[138],"real-world":[140],"experiment":[141],"using":[142],"8-node":[144],"50Gb":[148],"RDMA":[149],"four":[152],"representative":[153],"DNN":[154],"models.":[155],"Results":[156],"show":[157],"our":[159,178],"method":[160,179],"time":[163],"10%-20%":[165],"compared":[166,171],"Horovod-NCCL,":[168],"6%-13%":[170],"tensor-fusion-optimization-only":[173],"methods.":[174],"Simulation":[175],"shows":[176],"achieves":[180],"best":[182],"scalability":[183],"speedup":[187],"1.2x":[189],"over":[190],"best-performing":[192],"baseline":[193],"as":[194],"scale":[196],"up":[197],"size.":[199]},"counts_by_year":[{"year":2025,"cited_by_count":1},{"year":2024,"cited_by_count":1},{"year":2023,"cited_by_count":1}],"updated_date":"2025-12-24T23:09:58.560324","created_date":"2025-10-10T00:00:00"}