{"id":"https://openalex.org/W7163136911","doi":"https://doi.org/10.48550/arxiv.2606.00946","title":"Lodestar: An Online-Learning LLM Inference Router","display_name":"Lodestar: An Online-Learning LLM Inference Router","publication_year":2026,"publication_date":"2026-05-31","ids":{"openalex":"https://openalex.org/W7163136911","doi":"https://doi.org/10.48550/arxiv.2606.00946"},"language":null,"primary_location":{"id":"doi:10.48550/arxiv.2606.00946","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2606.00946","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.2606.00946","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5023957127","display_name":"Gangmuk Lim","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Lim, Gangmuk","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5137673056","display_name":"Wanyu Zhao","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zhao, Wanyu","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5082789555","display_name":"Brighten Godfrey","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Godfrey, Brighten","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5137710487","display_name":"Jiaxin Shan","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Shan, Jiaxin","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5137616029","display_name":"Le Xu","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Xu, Le","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5040158607","display_name":"Liguang Xie","orcid":"https://orcid.org/0000-0002-3764-9888"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Xie, Liguang","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":0.33219999074935913,"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":0.33219999074935913,"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/T12127","display_name":"Software System Performance and Reliability","score":0.08609999716281891,"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":0.08330000191926956,"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/latency","display_name":"Latency (audio)","score":0.5514000058174133},{"id":"https://openalex.org/keywords/inference","display_name":"Inference","score":0.5468999743461609},{"id":"https://openalex.org/keywords/cache","display_name":"Cache","score":0.5116999745368958},{"id":"https://openalex.org/keywords/router","display_name":"Router","score":0.4765999913215637},{"id":"https://openalex.org/keywords/cloud-computing","display_name":"Cloud computing","score":0.4652999937534332},{"id":"https://openalex.org/keywords/reuse","display_name":"Reuse","score":0.46299999952316284},{"id":"https://openalex.org/keywords/heuristics","display_name":"Heuristics","score":0.43720000982284546},{"id":"https://openalex.org/keywords/routing-table","display_name":"Routing table","score":0.4341000020503998},{"id":"https://openalex.org/keywords/snapshot","display_name":"Snapshot (computer storage)","score":0.42660000920295715}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.8819000124931335},{"id":"https://openalex.org/C82876162","wikidata":"https://www.wikidata.org/wiki/Q17096504","display_name":"Latency (audio)","level":2,"score":0.5514000058174133},{"id":"https://openalex.org/C2776214188","wikidata":"https://www.wikidata.org/wiki/Q408386","display_name":"Inference","level":2,"score":0.5468999743461609},{"id":"https://openalex.org/C115537543","wikidata":"https://www.wikidata.org/wiki/Q165596","display_name":"Cache","level":2,"score":0.5116999745368958},{"id":"https://openalex.org/C120314980","wikidata":"https://www.wikidata.org/wiki/Q180634","display_name":"Distributed computing","level":1,"score":0.5034000277519226},{"id":"https://openalex.org/C2775896111","wikidata":"https://www.wikidata.org/wiki/Q642560","display_name":"Router","level":2,"score":0.4765999913215637},{"id":"https://openalex.org/C79974875","wikidata":"https://www.wikidata.org/wiki/Q483639","display_name":"Cloud computing","level":2,"score":0.4652999937534332},{"id":"https://openalex.org/C206588197","wikidata":"https://www.wikidata.org/wiki/Q846574","display_name":"Reuse","level":2,"score":0.46299999952316284},{"id":"https://openalex.org/C173608175","wikidata":"https://www.wikidata.org/wiki/Q232661","display_name":"Parallel computing","level":1,"score":0.4424999952316284},{"id":"https://openalex.org/C127705205","wikidata":"https://www.wikidata.org/wiki/Q5748245","display_name":"Heuristics","level":2,"score":0.43720000982284546},{"id":"https://openalex.org/C184896649","wikidata":"https://www.wikidata.org/wiki/Q290066","display_name":"Routing table","level":4,"score":0.4341000020503998},{"id":"https://openalex.org/C55282118","wikidata":"https://www.wikidata.org/wiki/Q252683","display_name":"Snapshot (computer storage)","level":2,"score":0.42660000920295715},{"id":"https://openalex.org/C53833338","wikidata":"https://www.wikidata.org/wiki/Q1061424","display_name":"Context switch","level":2,"score":0.375900000333786},{"id":"https://openalex.org/C31258907","wikidata":"https://www.wikidata.org/wiki/Q1301371","display_name":"Computer network","level":1,"score":0.374099999666214},{"id":"https://openalex.org/C2779343474","wikidata":"https://www.wikidata.org/wiki/Q3109175","display_name":"Context (archaeology)","level":2,"score":0.33629998564720154},{"id":"https://openalex.org/C138959212","wikidata":"https://www.wikidata.org/wiki/Q1806783","display_name":"Load balancing (electrical power)","level":3,"score":0.3287999927997589},{"id":"https://openalex.org/C66882249","wikidata":"https://www.wikidata.org/wiki/Q169336","display_name":"Homogeneous","level":2,"score":0.3278000056743622},{"id":"https://openalex.org/C74172769","wikidata":"https://www.wikidata.org/wiki/Q1446839","display_name":"Routing (electronic design automation)","level":2,"score":0.3278000056743622},{"id":"https://openalex.org/C206729178","wikidata":"https://www.wikidata.org/wiki/Q2271896","display_name":"Scheduling (production processes)","level":2,"score":0.3172000050544739},{"id":"https://openalex.org/C190290938","wikidata":"https://www.wikidata.org/wiki/Q387015","display_name":"Trie","level":3,"score":0.31709998846054077},{"id":"https://openalex.org/C2984173633","wikidata":"https://www.wikidata.org/wiki/Q22725","display_name":"Routing algorithm","level":4,"score":0.3061000108718872},{"id":"https://openalex.org/C158379750","wikidata":"https://www.wikidata.org/wiki/Q214111","display_name":"Network packet","level":2,"score":0.304500013589859},{"id":"https://openalex.org/C17349429","wikidata":"https://www.wikidata.org/wiki/Q1049914","display_name":"Matrix multiplication","level":3,"score":0.29980000853538513},{"id":"https://openalex.org/C67388219","wikidata":"https://www.wikidata.org/wiki/Q207440","display_name":"Hash table","level":3,"score":0.298799991607666},{"id":"https://openalex.org/C2989134064","wikidata":"https://www.wikidata.org/wiki/Q288510","display_name":"Execution time","level":2,"score":0.29760000109672546},{"id":"https://openalex.org/C156884757","wikidata":"https://www.wikidata.org/wiki/Q798554","display_name":"Backtracking","level":2,"score":0.28450000286102295},{"id":"https://openalex.org/C187691185","wikidata":"https://www.wikidata.org/wiki/Q2020720","display_name":"Grid","level":2,"score":0.2799000144004822},{"id":"https://openalex.org/C79403827","wikidata":"https://www.wikidata.org/wiki/Q3988","display_name":"Real-time computing","level":1,"score":0.2533000111579895}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2606.00946","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2606.00946","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.2606.00946","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2606.00946","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":[{"display_name":"Industry, innovation and infrastructure","score":0.6544977426528931,"id":"https://metadata.un.org/sdg/9"}],"awards":[],"funders":[],"has_content":{"pdf":false,"grobid_xml":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"Efficiently":[0],"serving":[1,148],"large":[2],"language":[3],"model":[4],"(LLM)":[5],"inference":[6,30,127],"tasks":[7],"is":[8,36,40,141],"crucial":[9],"both":[10],"for":[11,19,75,92],"user-perceived":[12],"latency":[13,52],"such":[14],"as":[15],"time-to-first-token":[16],"(TTFT)":[17],"and":[18,44,51,60,71,113,116,143,158,167,179,190,193],"GPU":[20,34,94,210],"utilization.":[21],"However,":[22],"LLM":[23,76],"request":[24,31,89,111],"routing,":[25],"that":[26,122,132],"is,":[27],"assigning":[28],"each":[29],"to":[32,55,79,125,129,155,175,185],"a":[33,64,86,99,186,207],"instance,":[35],"particularly":[37],"challenging:":[38],"execution":[39],"highly":[41],"input-dependent;":[42],"batching":[43],"KV-cache":[45],"reuse":[46],"create":[47],"strong":[48],"cross-request":[49],"coupling;":[50],"responds":[53],"nonlinearly":[54],"context":[56],"length,":[57],"model/engine":[58],"settings,":[59],"heterogeneous":[61,182],"accelerators.":[62],"As":[63],"result,":[65],"simple":[66],"traditional":[67],"load":[68],"balancing":[69],"algorithms,":[70],"even":[72],"heuristics":[73],"tailored":[74],"inference,":[77],"fail":[78],"achieve":[80],"good":[81],"performance.":[82],"We":[83],"present":[84],"Lodestar,":[85],"novel":[87],"learning-based":[88],"routing":[90,197],"system":[91],"distributed":[93],"clusters.":[95],"Lodestar":[96,140,161],"continuously":[97],"collects":[98],"snapshot":[100],"of":[101],"the":[102,130],"cluster":[103],"at":[104],"per-request":[105],"level,":[106],"including":[107],"real-time":[108],"instance":[109,131],"state,":[110],"characteristics,":[112],"observed":[114],"performance,":[115],"trains":[117],"an":[118],"online":[119,153],"reward":[120,136],"predictor":[121],"it":[123],"uses":[124],"route":[126],"requests":[128],"will":[133],"maximize":[134],"given":[135],"(e.g.,":[137],"minimizing":[138],"TTFT).":[139],"cloud-native":[142],"works":[144],"seamlessly":[145],"with":[146],"existing":[147],"stacks":[149],"(vLLM).":[150],"With":[151],"continuous":[152],"adaptation":[154],"changing":[156],"workloads":[157],"infrastructure":[159],"conditions,":[160],"achieves":[162],"1.41x":[163],"lower":[164,169],"average":[165,173],"TTFT":[166,171],"1.47x":[168],"P99":[170],"on":[172,177,181,204],"(up":[174],"2.15x/1.86x":[176],"homogeneous":[178],"4.38x/4.42x":[180],"clusters)":[183],"compared":[184],"state-of-the-art":[187],"prefix":[188],"cache":[189],"load-aware":[191],"heuristic,":[192],"learns":[194],"these":[195],"efficient":[196],"strategies":[198],"within":[199],"about":[200],"5":[201],"minutes,":[202],"based":[203],"experiments":[205],"in":[206],"public":[208],"cloud":[209],"cluster.":[211]},"counts_by_year":[],"updated_date":"2026-07-01T06:00:48.157686","created_date":"2026-06-03T00:00:00"}