{"id":"https://openalex.org/W7225430076","doi":"https://doi.org/10.5281/zenodo.24010102","title":"The exact worst case of the global policy of DeepSeek's expert-parallel load balancer","display_name":"The exact worst case of the global policy of DeepSeek's expert-parallel load balancer","publication_year":2026,"publication_date":"2026-11-06","ids":{"openalex":"https://openalex.org/W7225430076","doi":"https://doi.org/10.5281/zenodo.24010102"},"language":"en","primary_location":{"id":"doi:10.5281/zenodo.24010102","doi":"https://doi.org/10.5281/zenodo.24010102","is_oa":true,"landing_page_url":"https://doi.org/10.5281/zenodo.24010102","pdf_url":null,"source":{"id":"https://openalex.org/S4306400562","display_name":"Zenodo (CERN European Organization for Nuclear Research)","issn_l":null,"issn":null,"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I67311998","host_organization_name":"European Organization for Nuclear Research","host_organization_lineage":["https://openalex.org/I67311998"],"host_organization_lineage_names":[],"type":"repository"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":null,"is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"Preprint"},"volume":null,"type":"preprint","is_traditional":true,"indexed_in":["datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://doi.org/10.5281/zenodo.24010102","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5119746719","display_name":"Kundan Baliga","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I157725225","display_name":"University of Illinois Urbana-Champaign","ror":"https://ror.org/047426m28","country_code":"US","type":"education","lineage":["https://openalex.org/I157725225"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Baliga, Kundan","raw_affiliation_strings":["University of Illinois Urbana-Champaign"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"University of Illinois Urbana-Champaign","institution_ids":["https://openalex.org/I157725225"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5160685850","display_name":"Aiden Xie","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I130701444","display_name":"Georgia Institute of Technology","ror":"https://ror.org/01zkghx44","country_code":"US","type":"education","lineage":["https://openalex.org/I130701444"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Xie, Aiden","raw_affiliation_strings":["Georgia Institute of Technology"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Georgia Institute of Technology","institution_ids":["https://openalex.org/I130701444"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5160585827","display_name":"Aditya Byju","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I130701444","display_name":"Georgia Institute of Technology","ror":"https://ror.org/01zkghx44","country_code":"US","type":"education","lineage":["https://openalex.org/I130701444"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Byju, Aditya","raw_affiliation_strings":["Georgia Institute of Technology"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Georgia Institute of Technology","institution_ids":["https://openalex.org/I130701444"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":2,"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/T10715","display_name":"Distributed and Parallel Computing Systems","score":0.4032999873161316,"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/T10715","display_name":"Distributed and Parallel Computing Systems","score":0.4032999873161316,"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":0.2824999988079071,"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"}},{"id":"https://openalex.org/T10054","display_name":"Parallel Computing and Optimization Techniques","score":0.19920000433921814,"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/deepseek","display_name":"DeepSeek","score":0.7329999804496765},{"id":"https://openalex.org/keywords/load-balancing","display_name":"load balancing","score":0.5509999990463257},{"id":"https://openalex.org/keywords/worst-case-analysis","display_name":"worst-case analysis","score":0.6819999814033508},{"id":"https://openalex.org/keywords/approximation-ratio","display_name":"approximation ratio","score":0.7480000257492065},{"id":"https://openalex.org/keywords/top-k-routing","display_name":"top-k routing","score":0.7289999723434448},{"id":"https://openalex.org/keywords/exchange-algorithm","display_name":"exchange algorithm","score":0.44200000166893005},{"id":"https://openalex.org/keywords/approximation-lower-bounds","display_name":"approximation lower bounds","score":0.35600000619888306},{"id":"https://openalex.org/keywords/large-language-models","display_name":"large language models","score":0.4959999918937683},{"id":"https://openalex.org/keywords/hierarchical-load-balancing","display_name":"hierarchical load balancing","score":0.38499999046325684}],"concepts":[{"id":"https://openalex.org/C77553402","wikidata":"https://www.wikidata.org/wiki/Q13222579","display_name":"Upper and lower bounds","level":2,"score":0.5785999894142151},{"id":"https://openalex.org/C151201525","wikidata":"https://www.wikidata.org/wiki/Q177239","display_name":"Limit (mathematics)","level":2,"score":0.532800018787384},{"id":"https://openalex.org/C2775937380","wikidata":"https://www.wikidata.org/wiki/Q1232589","display_name":"Replica","level":2,"score":0.5202000141143799},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.5192999839782715},{"id":"https://openalex.org/C74172769","wikidata":"https://www.wikidata.org/wiki/Q1446839","display_name":"Routing (electronic design automation)","level":2,"score":0.4666999876499176},{"id":"https://openalex.org/C126255220","wikidata":"https://www.wikidata.org/wiki/Q141495","display_name":"Mathematical optimization","level":1,"score":0.4555000066757202},{"id":"https://openalex.org/C18555067","wikidata":"https://www.wikidata.org/wiki/Q8375051","display_name":"Joint (building)","level":2,"score":0.4092000126838684},{"id":"https://openalex.org/C2781039887","wikidata":"https://www.wikidata.org/wiki/Q1391724","display_name":"Factor (programming language)","level":2,"score":0.39169999957084656},{"id":"https://openalex.org/C156340839","wikidata":"https://www.wikidata.org/wiki/Q2704791","display_name":"Enumeration","level":2,"score":0.36039999127388},{"id":"https://openalex.org/C33923547","wikidata":"https://www.wikidata.org/wiki/Q395","display_name":"Mathematics","level":0,"score":0.3578999936580658},{"id":"https://openalex.org/C2776505523","wikidata":"https://www.wikidata.org/wiki/Q4785468","display_name":"Plan (archaeology)","level":2,"score":0.35010001063346863},{"id":"https://openalex.org/C2778337684","wikidata":"https://www.wikidata.org/wiki/Q3173700","display_name":"Apportionment","level":2,"score":0.3424000144004822},{"id":"https://openalex.org/C93693863","wikidata":"https://www.wikidata.org/wiki/Q897659","display_name":"Branch and bound","level":2,"score":0.2727000117301941},{"id":"https://openalex.org/C167927819","wikidata":"https://www.wikidata.org/wiki/Q1930567","display_name":"Shuffling","level":2,"score":0.2524999976158142}],"mesh":[],"locations_count":5,"locations":[{"id":"doi:10.5281/zenodo.24010102","doi":"https://doi.org/10.5281/zenodo.24010102","is_oa":true,"landing_page_url":"https://doi.org/10.5281/zenodo.24010102","pdf_url":null,"source":{"id":"https://openalex.org/S4306400562","display_name":"Zenodo (CERN European Organization for Nuclear Research)","issn_l":null,"issn":null,"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I67311998","host_organization_name":"European Organization for Nuclear Research","host_organization_lineage":["https://openalex.org/I67311998"],"host_organization_lineage_names":[],"type":"repository"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":null,"is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"Preprint"},{"id":"doi:10.5281/zenodo.24010103","doi":"https://doi.org/10.5281/zenodo.24010103","is_oa":true,"landing_page_url":"https://doi.org/10.5281/zenodo.24010103","pdf_url":null,"source":{"id":"https://openalex.org/S4306400562","display_name":"Zenodo (CERN European Organization for Nuclear Research)","issn_l":null,"issn":null,"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I67311998","host_organization_name":"European Organization for Nuclear Research","host_organization_lineage":["https://openalex.org/I67311998"],"host_organization_lineage_names":[],"type":"repository"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":null,"is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"Preprint"},{"id":"doi:10.5281/zenodo.24014217","doi":"https://doi.org/10.5281/zenodo.24014217","is_oa":true,"landing_page_url":"https://doi.org/10.5281/zenodo.24014217","pdf_url":null,"source":{"id":"https://openalex.org/S4306400562","display_name":"Zenodo (CERN European Organization for Nuclear Research)","issn_l":null,"issn":null,"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I67311998","host_organization_name":"European Organization for Nuclear Research","host_organization_lineage":["https://openalex.org/I67311998"],"host_organization_lineage_names":[],"type":"repository"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":null,"is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"Preprint"},{"id":"doi:10.5281/zenodo.24015151","doi":"https://doi.org/10.5281/zenodo.24015151","is_oa":true,"landing_page_url":"https://doi.org/10.5281/zenodo.24015151","pdf_url":null,"source":{"id":"https://openalex.org/S4306400562","display_name":"Zenodo (CERN European Organization for Nuclear Research)","issn_l":null,"issn":null,"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I67311998","host_organization_name":"European Organization for Nuclear Research","host_organization_lineage":["https://openalex.org/I67311998"],"host_organization_lineage_names":[],"type":"repository"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":null,"is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"Preprint"},{"id":"doi:10.5281/zenodo.24021645","doi":"https://doi.org/10.5281/zenodo.24021645","is_oa":true,"landing_page_url":"https://doi.org/10.5281/zenodo.24021645","pdf_url":null,"source":{"id":"https://openalex.org/S4306400562","display_name":"Zenodo (CERN European Organization for Nuclear Research)","issn_l":null,"issn":null,"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I67311998","host_organization_name":"European Organization for Nuclear Research","host_organization_lineage":["https://openalex.org/I67311998"],"host_organization_lineage_names":[],"type":"repository"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":null,"is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"Preprint"}],"best_oa_location":{"id":"doi:10.5281/zenodo.24010102","doi":"https://doi.org/10.5281/zenodo.24010102","is_oa":true,"landing_page_url":"https://doi.org/10.5281/zenodo.24010102","pdf_url":null,"source":{"id":"https://openalex.org/S4306400562","display_name":"Zenodo (CERN European Organization for Nuclear Research)","issn_l":null,"issn":null,"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I67311998","host_organization_name":"European Organization for Nuclear Research","host_organization_lineage":["https://openalex.org/I67311998"],"host_organization_lineage_names":[],"type":"repository"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":null,"is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"Preprint"},"sustainable_development_goals":[],"study_designs":[],"awards":[],"funders":[],"has_content":{"pdf":false,"grobid_xml":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":["https://openalex.org/W7203480443","https://openalex.org/W7214712040","https://openalex.org/W7155163985","https://openalex.org/W7125796593","https://openalex.org/W7163556042","https://openalex.org/W6987595827","https://openalex.org/W7147577857","https://openalex.org/W4417290844","https://openalex.org/W7115041457","https://openalex.org/W7216508552"],"abstract_inverted_index":{"DeepSeek's":[0,181],"expert-parallel":[1],"load":[2,28,136,277,285,410],"balancer":[3],"(EPLB)":[4],"replicates":[5],"experts":[6,146],"by":[7,18,227,390],"Adams's":[8],"apportionment":[9],"method":[10],"and":[11,62,111,137,191,212,230,253,264,287,308,327,343,350,393,406,421,465],"then":[12],"packs":[13],"the":[14,55,94,106,113,121,133,139,151,164,209,213,223,271,274,279,284,288,300,309,313,321,336,340,347,359,365,394,432,439,444,457,467,474],"replicas":[15],"onto":[16],"GPUs":[17],"a":[19,91,98,169,217,254],"cardinality-constrained":[20],"longest-processing-time":[21],"rule":[22],"(modified":[23],"LPT).":[24],"In":[25],"EPLB's":[26],"own":[27],"model":[29,286],"(static":[30],"expert":[31,117,430],"loads,":[32],"each":[33],"expert's":[34],"tokens":[35],"split":[36],"evenly":[37],"among":[38],"its":[39,44,265],"copies)":[40],"we":[41],"prove":[42],"that":[43],"global":[45,210],"policy":[46,211,215,477],"has":[47],"worst-case":[48],"maximum-GPU-load":[49],"ratio":[50,334],"exactly":[51],"2m/(m+1)":[52],"relative":[53],"to":[54,292,470,480],"best":[56],"joint":[57],"choice":[58],"of":[59,93,105,324,428,456],"replica":[60],"counts":[61],"placement":[63],"(m":[64],"GPUs;":[65],"attained":[66],"whenever":[67],"there":[68],"are":[69,267,345,485],"at":[70,102,158,185,199],"least":[71,103,186],"two":[72],"slots":[73],"per":[74,82],"GPU;":[75],"EPLB":[76,422],"is":[77,132,173,295,335,354,371,396,478],"optimal":[78,107],"with":[79,320,443],"one":[80,116,429],"slot":[81],"GPU).":[83],"The":[84,206,244],"worst":[85,166],"case":[86,167],"needs":[87],"very":[88],"concentrated":[89],"load:":[90],"loss":[92],"full":[95,314],"factor":[96,366,395],"requires":[97],"single":[99,218],"copy":[100,135],"carrying":[101],"m/(m+1)":[104],"maximum":[108],"GPU":[109,434],"load,":[110],"in":[112,235,259,358],"bad":[114],"instances":[115],"carries":[118],"almost":[119],"all":[120,247,452],"traffic.":[122],"More":[123],"generally":[124],"EPLB/OPT":[125,303],"\u2264":[126,304],"1":[127,305],"+":[128,306],"(m-1)M/T,":[129,307],"where":[130],"M":[131],"largest":[134],"T":[138],"total":[140],"load.":[141],"With":[142],"K":[143],"equally":[144],"hot":[145,189],"(as":[147],"top-K":[148],"routing":[149],"suggests)":[150],"same":[152,337,433],"construction":[153],"gives":[154,184],"2m/(m+K),":[155],"e.g.":[156],"1.6":[157],"K=8,":[159],"m=32":[160],"(a":[161],"lower":[162],"bound;":[163],"exact":[165],"under":[168,438],"1/K":[170],"share":[171],"limit":[172],"open).":[174],"On":[175],"constructed":[176,488],"inputs":[177],"(not":[178],"measured":[179],"traffic),":[180],"released":[182,445],"eplb.py":[183],"1.9391":[187],"(one":[188],"expert)":[190],"1.5999":[192],"(eight)":[193],"times":[194],"an":[195,250,471],"explicit":[196],"better":[197],"plan":[198],"256":[200],"experts,":[201],"32":[202],"GPUs,":[203],"9":[204],"slots.":[205],"analysis":[207],"covers":[208],"hierarchical":[214,440,476],"on":[216,364,414,431,435],"node.":[219],"An":[220],"appendix":[221],"certifies":[222],"(4m-1)/(3m)":[224],"bound":[225,302],"conjectured":[226],"Babel,":[228,325],"Kellerer":[229,326],"Kotov":[231,328],"for":[232,240,312,368,412,451,463],"modified":[233,369],"LPT":[234,370],"14":[236,453],"small":[237],"(m,k)":[238,454],"cases,":[239],"every":[241],"tie-breaking":[242],"rule.":[243],"supplement":[245],"contains":[246],"code,":[248],"certificates,":[249],"independent":[251],"checker":[252],"README":[255],"(MIT":[256],"license).":[257],"Changes":[258],"v1.1":[260],"(the":[261,378],"main":[262],"theorem":[263],"proof":[266,351],"unchanged):":[268],"Title":[269],"adds":[270],"qualifier":[272],"\"in":[273],"static,":[275],"even-split":[276],"model\";":[278],"abstract's":[280],"first":[281],"paragraph":[282],"states":[283],"concentrated-load":[289],"requirement,":[290],"worded":[291],"match":[293],"what":[294,374],"proven.":[296],"New":[297,316,401,447],"Corollary":[298],"5:":[299],"instance-wise":[301],"heavy-copy":[310],"condition":[311],"factor.":[315],"Remark":[317],"7":[318],"comparing":[319],"exchange":[322],"algorithm":[323,349],"(1998,":[329],"Theorem":[330],"6),":[331],"whose":[332],"tight":[333],"constant":[338],"2-2/(m+1):":[339],"extremal":[341],"shape":[342],"accounting":[344],"shared;":[346],"items,":[348],"differ.":[352],"Babel-Kellerer-Kotov":[353],"now":[355],"introduced":[356],"once":[357],"related":[360,402],"work.":[361],"Prior":[362],"work":[363],"2-1/m":[367],"described":[372],"from":[373],"could":[375],"be":[376,481],"read":[377],"Dell'Amico-Martello":[379],"2001":[380],"abstract;":[381],"Wu-Yao":[382],"2008,":[383],"Appl.":[384],"Math.":[385],"J.":[386],"Chinese":[387],"Univ.":[388],"23:1-8,":[389],"title":[391],"only)":[392],"not":[397],"claimed":[398],"as":[399,487],"new.":[400],"work:":[403],"Wolf,":[404],"Yu":[405],"Shachnai's":[407],"replicate-then-place":[408],"disk":[409],"balancing":[411],"video":[413],"demand":[415],"(SIGMETRICS":[416],"1995;":[417],"Multimedia":[418],"Systems":[419],"1997),":[420],"GitHub":[423],"issue":[424],"#22":[425],"(two":[426],"copies":[427],"real":[436],"traces,":[437],"policy;":[441],"reproduced":[442],"code).":[446],"computation:":[448],"every-tie-rule":[449],"certificates":[450],"cases":[455],"conjecture":[458],"(v1.0":[459],"had":[460],"them":[461],"only":[462],"(2,4)":[464],"(3,4));":[466],"certification":[468],"moved":[469],"appendix.":[472],"Scope:":[473],"single-node":[475],"shown":[479],"covered;":[482],"real-code":[483],"ratios":[484],"labelled":[486],"instances.":[489]},"counts_by_year":[],"updated_date":"2026-11-08T12:12:28.789561","created_date":"2026-11-07T00:00:00"}
