{"id":"https://openalex.org/W7225459072","doi":"https://doi.org/10.5281/zenodo.24023728","title":"The global policy of DeepSeek's expert-parallel load balancer is a tight 2m/(m+1)-approximation for static, evenly split loads","display_name":"The global policy of DeepSeek's expert-parallel load balancer is a tight 2m/(m+1)-approximation for static, evenly split loads","publication_year":2026,"publication_date":"2026-11-06","ids":{"openalex":"https://openalex.org/W7225459072","doi":"https://doi.org/10.5281/zenodo.24023728"},"language":"en","primary_location":{"id":"doi:10.5281/zenodo.24023728","doi":"https://doi.org/10.5281/zenodo.24023728","is_oa":true,"landing_page_url":"https://doi.org/10.5281/zenodo.24023728","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.24023728","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/T10551","display_name":"Scheduling and Optimization Algorithms","score":0.2605000138282776,"subfield":{"id":"https://openalex.org/subfields/2209","display_name":"Industrial and Manufacturing Engineering"},"field":{"id":"https://openalex.org/fields/22","display_name":"Engineering"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},"topics":[{"id":"https://openalex.org/T10551","display_name":"Scheduling and Optimization Algorithms","score":0.2605000138282776,"subfield":{"id":"https://openalex.org/subfields/2209","display_name":"Industrial and Manufacturing Engineering"},"field":{"id":"https://openalex.org/fields/22","display_name":"Engineering"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},{"id":"https://openalex.org/T12176","display_name":"Optimization and Packing Problems","score":0.1492999941110611,"subfield":{"id":"https://openalex.org/subfields/2209","display_name":"Industrial and Manufacturing Engineering"},"field":{"id":"https://openalex.org/fields/22","display_name":"Engineering"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},{"id":"https://openalex.org/T10715","display_name":"Distributed and Parallel Computing Systems","score":0.11420000344514847,"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"}}],"keywords":[{"id":"https://openalex.org/keywords/mixture-of-experts","display_name":"mixture of experts","score":0.7269999980926514},{"id":"https://openalex.org/keywords/load-balancing","display_name":"load balancing","score":0.7409999966621399},{"id":"https://openalex.org/keywords/greedy-scheduling","display_name":"greedy scheduling","score":0.5540000200271606},{"id":"https://openalex.org/keywords/approximation-ratio","display_name":"approximation ratio","score":0.8050000071525574},{"id":"https://openalex.org/keywords/gpu-scheduling","display_name":"GPU scheduling","score":0.5059999823570251},{"id":"https://openalex.org/keywords/approximation-guarantees","display_name":"approximation guarantees","score":0.3019999861717224}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.6818000078201294},{"id":"https://openalex.org/C206729178","wikidata":"https://www.wikidata.org/wiki/Q2271896","display_name":"Scheduling (production processes)","level":2,"score":0.6007999777793884},{"id":"https://openalex.org/C194648553","wikidata":"https://www.wikidata.org/wiki/Q1364774","display_name":"Spare part","level":2,"score":0.5490999817848206},{"id":"https://openalex.org/C58328972","wikidata":"https://www.wikidata.org/wiki/Q184609","display_name":"Expert system","level":2,"score":0.5282999873161316},{"id":"https://openalex.org/C2776760102","wikidata":"https://www.wikidata.org/wiki/Q5139990","display_name":"Code (set theory)","level":3,"score":0.4625000059604645},{"id":"https://openalex.org/C2781039887","wikidata":"https://www.wikidata.org/wiki/Q1391724","display_name":"Factor (programming language)","level":2,"score":0.42829999327659607},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.3643999993801117},{"id":"https://openalex.org/C2164484","wikidata":"https://www.wikidata.org/wiki/Q5170150","display_name":"Core (optical fiber)","level":2,"score":0.3443000018596649},{"id":"https://openalex.org/C138959212","wikidata":"https://www.wikidata.org/wiki/Q1806783","display_name":"Load balancing (electrical power)","level":3,"score":0.3384999930858612},{"id":"https://openalex.org/C126255220","wikidata":"https://www.wikidata.org/wiki/Q141495","display_name":"Mathematical optimization","level":1,"score":0.33660000562667847},{"id":"https://openalex.org/C2984822820","wikidata":"https://www.wikidata.org/wiki/Q1123036","display_name":"Processor scheduling","level":3,"score":0.3301999866962433},{"id":"https://openalex.org/C173608175","wikidata":"https://www.wikidata.org/wiki/Q232661","display_name":"Parallel computing","level":1,"score":0.301800012588501},{"id":"https://openalex.org/C3283095","wikidata":"https://www.wikidata.org/wiki/Q1340282","display_name":"Load factor","level":2,"score":0.29679998755455017},{"id":"https://openalex.org/C176553487","wikidata":"https://www.wikidata.org/wiki/Q7855819","display_name":"Turnaround time","level":2,"score":0.2782000005245209},{"id":"https://openalex.org/C55416958","wikidata":"https://www.wikidata.org/wiki/Q6206757","display_name":"Job shop scheduling","level":3,"score":0.2720000147819519},{"id":"https://openalex.org/C111873713","wikidata":"https://www.wikidata.org/wiki/Q1641413","display_name":"Job scheduler","level":3,"score":0.2628999948501587},{"id":"https://openalex.org/C2781172179","wikidata":"https://www.wikidata.org/wiki/Q853109","display_name":"Parallelism (grammar)","level":2,"score":0.2524000108242035},{"id":"https://openalex.org/C2989134064","wikidata":"https://www.wikidata.org/wiki/Q288510","display_name":"Execution time","level":2,"score":0.2508000135421753}],"mesh":[],"locations_count":3,"locations":[{"id":"doi:10.5281/zenodo.24023728","doi":"https://doi.org/10.5281/zenodo.24023728","is_oa":true,"landing_page_url":"https://doi.org/10.5281/zenodo.24023728","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.24023729","doi":"https://doi.org/10.5281/zenodo.24023729","is_oa":true,"landing_page_url":"https://doi.org/10.5281/zenodo.24023729","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.24059219","doi":"https://doi.org/10.5281/zenodo.24059219","is_oa":true,"landing_page_url":"https://doi.org/10.5281/zenodo.24059219","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.24023728","doi":"https://doi.org/10.5281/zenodo.24023728","is_oa":true,"landing_page_url":"https://doi.org/10.5281/zenodo.24023728","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":{"grobid_xml":false,"pdf":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":["https://openalex.org/W7225430076","https://openalex.org/W7203480443","https://openalex.org/W7214712040","https://openalex.org/W7155163985","https://openalex.org/W7125796593","https://openalex.org/W7115041457","https://openalex.org/W7163556042","https://openalex.org/W7147577857","https://openalex.org/W4413726998","https://openalex.org/W6987595827"],"abstract_inverted_index":{"DeepSeek's":[0],"Expert":[1],"Parallelism":[2],"Load":[3],"Balancer":[4],"(EPLB),":[5],"also":[6],"used":[7],"in":[8,155],"vLLM":[9],"and":[10,27,41,88,110,147,149,180,183],"SGLang,":[11],"uses":[12],"two":[13,101,122],"greedy":[14],"steps":[15],"to":[16,25,29],"choose":[17],"how":[18],"many":[19],"copies":[20,154],"of":[21,145],"each":[22,96],"mixture-of-experts":[23],"expert":[24,62,109,117],"keep":[26],"where":[28],"put":[30],"them.":[31],"The":[32,103,127,172],"first":[33],"step":[34],"is":[35,44,106],"optimal":[36,157],"for":[37,134],"its":[38,71,177],"own":[39,58],"goal":[40],"the":[42,50,79,86,116,142,167,175,181],"second":[43],"a":[45,125,131],"standard":[46],"scheduling":[47],"rule,":[48],"but":[49],"combination":[51],"had":[52],"no":[53],"known":[54],"guarantee.":[55],"In":[56],"EPLB's":[57],"model,":[59],"with":[60],"static":[61],"loads":[63,78],"split":[64],"evenly":[65],"among":[66],"copies,":[67,120],"we":[68],"show":[69],"that":[70,89,185],"global":[72],"policy":[73],"on":[74],"m":[75],"GPUs":[76],"never":[77],"busiest":[80],"GPU":[81,97],"more":[82],"than":[83],"2m/(m+1)":[84],"times":[85],"optimum,":[87],"some":[90],"instance":[91],"attains":[92],"this":[93],"factor":[94,143],"whenever":[95],"has":[98],"at":[99],"least":[100],"slots.":[102],"bad":[104],"case":[105],"one":[107,111],"hot":[108,162],"spare":[112],"slot":[113],"per":[114],"GPU:":[115],"gets":[118],"m+1":[119],"so":[121],"must":[123],"share":[124],"GPU.":[126],"proof":[128],"starts":[129],"from":[130],"Graham-type":[132],"inequality":[133],"largest-first":[135],"packing,":[136],"which":[137],"alone":[138],"gives":[139],"2-1/m":[140],"(improving":[141],"2":[144],"Dell'Amico":[146],"Martello),":[148],"sharpens":[150],"it":[151],"by":[152],"counting":[153],"an":[156],"plan.":[158],"With":[159],"K":[160],"equally":[161],"experts":[163],"(K":[164],"dividing":[165],"m)":[166],"loss":[168],"can":[169],"approach":[170],"2m/(m+K).":[171],"record":[173],"includes":[174],"paper,":[176],"LaTeX":[178],"source,":[179],"code":[182],"certificates":[184],"reproduce":[186],"every":[187],"computed":[188],"result.":[189]},"counts_by_year":[],"updated_date":"2026-11-09T13:22:03.289127","created_date":"2026-11-08T00:00:00"}
