{"id":"https://openalex.org/W7163219657","doi":"https://doi.org/10.48550/arxiv.2606.01839","title":"Observation, Not Prediction: Conversation-Level Disaggregated Scheduling for Agentic Serving","display_name":"Observation, Not Prediction: Conversation-Level Disaggregated Scheduling for Agentic Serving","publication_year":2026,"publication_date":"2026-06-01","ids":{"openalex":"https://openalex.org/W7163219657","doi":"https://doi.org/10.48550/arxiv.2606.01839"},"language":null,"primary_location":{"id":"doi:10.48550/arxiv.2606.01839","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2606.01839","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"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":"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"},"type":"preprint","indexed_in":["datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://doi.org/10.48550/arxiv.2606.01839","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5080788209","display_name":"Jianru Ding","orcid":"https://orcid.org/0009-0007-5282-1379"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Ding, Jianru","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5086390985","display_name":"Ryien Hosseini","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Hosseini, Ryien","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5091134015","display_name":"Pouya Mahdi Gholami","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Gholami, Pouya Mahdi","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5100556351","display_name":"Mingyuan Xiang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Xiang, Mingyuan","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5137704076","display_name":"Henry Hoffmann","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Hoffmann, Henry","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":0.34299999475479126,"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":0.34299999475479126,"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/T14347","display_name":"Big Data and Digital Economy","score":0.06129999831318855,"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/T12808","display_name":"Ferroelectric and Negative Capacitance Devices","score":0.05169999971985817,"subfield":{"id":"https://openalex.org/subfields/2208","display_name":"Electrical and Electronic Engineering"},"field":{"id":"https://openalex.org/fields/22","display_name":"Engineering"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}}],"keywords":[{"id":"https://openalex.org/keywords/scheduling","display_name":"Scheduling (production processes)","score":0.7225000262260437},{"id":"https://openalex.org/keywords/workload","display_name":"Workload","score":0.5741000175476074},{"id":"https://openalex.org/keywords/conversation","display_name":"Conversation","score":0.546999990940094},{"id":"https://openalex.org/keywords/cache","display_name":"Cache","score":0.5249000191688538},{"id":"https://openalex.org/keywords/latency","display_name":"Latency (audio)","score":0.4668999910354614},{"id":"https://openalex.org/keywords/inference","display_name":"Inference","score":0.428600013256073},{"id":"https://openalex.org/keywords/fair-share-scheduling","display_name":"Fair-share scheduling","score":0.3700000047683716},{"id":"https://openalex.org/keywords/fixed-priority-pre-emptive-scheduling","display_name":"Fixed-priority pre-emptive scheduling","score":0.3433000147342682}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.7773000001907349},{"id":"https://openalex.org/C206729178","wikidata":"https://www.wikidata.org/wiki/Q2271896","display_name":"Scheduling (production processes)","level":2,"score":0.7225000262260437},{"id":"https://openalex.org/C2778476105","wikidata":"https://www.wikidata.org/wiki/Q628539","display_name":"Workload","level":2,"score":0.5741000175476074},{"id":"https://openalex.org/C2777200299","wikidata":"https://www.wikidata.org/wiki/Q52943","display_name":"Conversation","level":2,"score":0.546999990940094},{"id":"https://openalex.org/C115537543","wikidata":"https://www.wikidata.org/wiki/Q165596","display_name":"Cache","level":2,"score":0.5249000191688538},{"id":"https://openalex.org/C82876162","wikidata":"https://www.wikidata.org/wiki/Q17096504","display_name":"Latency (audio)","level":2,"score":0.4668999910354614},{"id":"https://openalex.org/C2776214188","wikidata":"https://www.wikidata.org/wiki/Q408386","display_name":"Inference","level":2,"score":0.428600013256073},{"id":"https://openalex.org/C120314980","wikidata":"https://www.wikidata.org/wiki/Q180634","display_name":"Distributed computing","level":1,"score":0.4205999970436096},{"id":"https://openalex.org/C79403827","wikidata":"https://www.wikidata.org/wiki/Q3988","display_name":"Real-time computing","level":1,"score":0.3736000061035156},{"id":"https://openalex.org/C31689143","wikidata":"https://www.wikidata.org/wiki/Q733809","display_name":"Fair-share scheduling","level":3,"score":0.3700000047683716},{"id":"https://openalex.org/C122141398","wikidata":"https://www.wikidata.org/wiki/Q5456330","display_name":"Fixed-priority pre-emptive scheduling","level":5,"score":0.3433000147342682},{"id":"https://openalex.org/C175893541","wikidata":"https://www.wikidata.org/wiki/Q1196582","display_name":"Round-robin scheduling","level":4,"score":0.3158999979496002},{"id":"https://openalex.org/C107568181","wikidata":"https://www.wikidata.org/wiki/Q5319000","display_name":"Dynamic priority scheduling","level":3,"score":0.3138999938964844},{"id":"https://openalex.org/C119948110","wikidata":"https://www.wikidata.org/wiki/Q7858726","display_name":"Two-level scheduling","level":4,"score":0.31029999256134033},{"id":"https://openalex.org/C2742236","wikidata":"https://www.wikidata.org/wiki/Q924713","display_name":"Efficient energy use","level":2,"score":0.3093000054359436},{"id":"https://openalex.org/C165696696","wikidata":"https://www.wikidata.org/wiki/Q11287","display_name":"Exploit","level":2,"score":0.3089999854564667},{"id":"https://openalex.org/C2780451532","wikidata":"https://www.wikidata.org/wiki/Q759676","display_name":"Task (project management)","level":2,"score":0.29319998621940613},{"id":"https://openalex.org/C57273362","wikidata":"https://www.wikidata.org/wiki/Q576722","display_name":"Decoding methods","level":2,"score":0.2809999883174896},{"id":"https://openalex.org/C2984822820","wikidata":"https://www.wikidata.org/wiki/Q1123036","display_name":"Processor scheduling","level":3,"score":0.2703999876976013},{"id":"https://openalex.org/C108650721","wikidata":"https://www.wikidata.org/wiki/Q1783253","display_name":"Counterfactual thinking","level":2,"score":0.26989999413490295},{"id":"https://openalex.org/C127456818","wikidata":"https://www.wikidata.org/wiki/Q238879","display_name":"Rate-monotonic scheduling","level":4,"score":0.25209999084472656}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2606.01839","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2606.01839","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"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":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":null,"is_accepted":false,"is_published":null,"raw_source_name":null,"raw_type":"Preprint"}],"best_oa_location":{"id":"doi:10.48550/arxiv.2606.01839","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2606.01839","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"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":"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":[{"display_name":"Affordable and clean energy","id":"https://metadata.un.org/sdg/7","score":0.9096192717552185}],"awards":[],"funders":[],"has_content":{"pdf":false,"grobid_xml":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"LLM-based":[0],"agents":[1],"resolve":[2],"a":[3,16,106,111,118,157,171,186,197,225],"user":[4],"task":[5,25],"through":[6],"many":[7],"turns":[8],"of":[9,182,196],"dependent":[10],"inference":[11],"and":[12,37,58,138,166,204,214],"tool":[13,56],"calls,":[14],"producing":[15],"workload":[17],"whose":[18],"total":[19],"cost":[20],"is":[21,83],"unknown":[22],"when":[23,66],"the":[24,31,34,52,67,72,86,90,93,97,100,124,127,134,153,161,168,217],"arrives.":[26],"Existing":[27],"multi-turn":[28],"systems":[29],"keep":[30],"turn":[32,39,98],"as":[33,126],"scheduling":[35,87,94,128],"unit":[36,95],"decide,":[38],"by":[40,85,116,202,208],"turn,":[41],"whether":[42],"to":[43,74,99,132,156,170],"disaggregate":[44],"prefill":[45,114,155],"from":[46,96],"decode.":[47],"That":[48],"decision":[49],"rests":[50],"on":[51,81],"turn's":[53],"decode":[54],"length,":[55],"behavior,":[57],"KV":[59,140,162],"growth,":[60],"quantities":[61],"that":[62],"are":[63],"not":[64,89],"observable":[65],"scheduler":[68],"must":[69],"act,":[70],"forcing":[71],"system":[73],"predict":[75],"them.":[76],"We":[77,145],"show":[78],"this":[79,147],"dependence":[80],"prediction":[82,188],"imposed":[84],"unit,":[88,129],"workload.":[91],"Raising":[92],"conversation":[101,125,169],"converts":[102],"turn-level":[103],"irregularity":[104],"into":[105],"stable,":[107],"two-phase":[108],"structure:":[109],"1)":[110],"compute-bound":[112],"turn-1":[113],"followed":[115],"2)":[117],"long,":[119],"memory-bound":[120],"tail.":[121],"Thus,":[122],"with":[123,178],"placement":[130],"reduces":[131,191],"reading":[133],"first-turn":[135,154],"input":[136],"length":[137],"per-decoder":[139],"occupancy,":[141],"both":[142],"directly":[143],"observable.":[144],"instantiate":[146],"principle":[148],"in":[149,228],"ConServe,":[150],"which":[151],"routes":[152],"high-throughput":[158],"prefiller,":[159],"transfers":[160],"cache":[163],"exactly":[164],"once,":[165],"pins":[167],"single":[172],"decoder":[173],"for":[174],"its":[175],"entire":[176],"tail,":[177],"no":[179],"learned":[180],"model":[181],"decode-side":[183],"cost.":[184],"Against":[185],"per-turn":[187],"baseline,":[189],"ConServe":[190],"p95":[192],"time-to-first-effective-token":[193],"(the":[194],"latency":[195],"conversation's":[198],"first":[199],"user-visible":[200],"output)":[201],"51.08%":[203],"improves":[205],"energy":[206,229],"efficiency":[207],"7.51%":[209],"while":[210],"preserving":[211],"last-turn":[212],"TBT":[213],"SLOs;":[215],"mapping":[216],"two":[218],"phases":[219],"onto":[220],"heterogeneous":[221],"GPU":[222],"tiers":[223],"adds":[224],"further":[226],"22.75%":[227],"efficiency.":[230]},"counts_by_year":[],"updated_date":"2026-07-28T07:46:37.118299","created_date":"2026-06-03T00:00:00"}