{"id":"https://openalex.org/W7214397909","doi":"https://doi.org/10.48550/arxiv.2609.28870","title":"When Fancy Eviction Fails: Rethinking Cache Replacement For LLM Prefix Reuse","display_name":"When Fancy Eviction Fails: Rethinking Cache Replacement For LLM Prefix Reuse","publication_year":2026,"publication_date":"2026-09-24","ids":{"openalex":"https://openalex.org/W7214397909","doi":"https://doi.org/10.48550/arxiv.2609.28870"},"language":"en","primary_location":{"id":"pmh:oai:arXiv.org:2609.28870","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2609.28870","pdf_url":"https://arxiv.org/pdf/2609.28870","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,"listed_in":[],"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":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},"volume":null,"type":"preprint","is_traditional":true,"indexed_in":["arxiv","datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://arxiv.org/pdf/2609.28870","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5124980337","display_name":"Yiyu Liu","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I136199984","display_name":"Harvard University","ror":"https://ror.org/03vek6s52","country_code":"US","type":"education","lineage":["https://openalex.org/I136199984"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Liu, Yiyu","raw_affiliation_strings":["Harvard University"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Harvard University","institution_ids":["https://openalex.org/I136199984"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5035157838","display_name":"Minlan Yu","orcid":"https://orcid.org/0000-0002-2381-0212","observed_orcids":["https://orcid.org/0000-0002-2381-0212"]},"institutions":[{"id":"https://openalex.org/I136199984","display_name":"Harvard University","ror":"https://ror.org/03vek6s52","country_code":"US","type":"education","lineage":["https://openalex.org/I136199984"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Yu, Minlan","raw_affiliation_strings":["Harvard University"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Harvard University","institution_ids":["https://openalex.org/I136199984"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5152790923","display_name":"Juncheng Yang","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I136199984","display_name":"Harvard University","ror":"https://ror.org/03vek6s52","country_code":"US","type":"education","lineage":["https://openalex.org/I136199984"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Yang, Juncheng","raw_affiliation_strings":["Harvard University"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Harvard University","institution_ids":["https://openalex.org/I136199984"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":1,"corresponding_author_ids":[],"corresponding_institution_ids":["https://openalex.org/I136199984"],"apc_list":null,"apc_paid":null,"fwci":null,"has_fulltext":true,"cited_by_count":3,"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":0.7764000296592712,"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":0.7764000296592712,"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/T11478","display_name":"Caching and Content Delivery","score":0.035599999129772186,"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/T12127","display_name":"Software System Performance and Reliability","score":0.032999999821186066,"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/lru","display_name":"LRU","score":0.5730000138282776}],"concepts":[{"id":"https://openalex.org/C141603448","wikidata":"https://www.wikidata.org/wiki/Q134830","display_name":"Prefix","level":2,"score":0.7670000195503235},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.766700029373169},{"id":"https://openalex.org/C2778227907","wikidata":"https://www.wikidata.org/wiki/Q1893186","display_name":"Eviction","level":2,"score":0.644599974155426},{"id":"https://openalex.org/C115537543","wikidata":"https://www.wikidata.org/wiki/Q165596","display_name":"Cache","level":2,"score":0.6419000029563904},{"id":"https://openalex.org/C2779182362","wikidata":"https://www.wikidata.org/wiki/Q17126187","display_name":"Session (web analytics)","level":2,"score":0.5645999908447266},{"id":"https://openalex.org/C206588197","wikidata":"https://www.wikidata.org/wiki/Q846574","display_name":"Reuse","level":2,"score":0.5156999826431274},{"id":"https://openalex.org/C2781092381","wikidata":"https://www.wikidata.org/wiki/Q5874546","display_name":"Hoarding (animal behavior)","level":3,"score":0.41589999198913574},{"id":"https://openalex.org/C77618280","wikidata":"https://www.wikidata.org/wiki/Q1155772","display_name":"Scheme (mathematics)","level":2,"score":0.4052000045776367},{"id":"https://openalex.org/C204495577","wikidata":"https://www.wikidata.org/wiki/Q1205349","display_name":"Callback","level":2,"score":0.4034999907016754},{"id":"https://openalex.org/C111919701","wikidata":"https://www.wikidata.org/wiki/Q9135","display_name":"Operating system","level":1,"score":0.35679998993873596},{"id":"https://openalex.org/C2778112365","wikidata":"https://www.wikidata.org/wiki/Q3511065","display_name":"Sequence (biology)","level":2,"score":0.3379000127315521},{"id":"https://openalex.org/C38652104","wikidata":"https://www.wikidata.org/wiki/Q3510521","display_name":"Computer security","level":1,"score":0.335099995136261},{"id":"https://openalex.org/C63584917","wikidata":"https://www.wikidata.org/wiki/Q333286","display_name":"Bounding overwatch","level":2,"score":0.32089999318122864},{"id":"https://openalex.org/C154504017","wikidata":"https://www.wikidata.org/wiki/Q853614","display_name":"Identifier","level":2,"score":0.2930000126361847},{"id":"https://openalex.org/C77088390","wikidata":"https://www.wikidata.org/wiki/Q8513","display_name":"Database","level":1,"score":0.29280000925064087},{"id":"https://openalex.org/C2778348673","wikidata":"https://www.wikidata.org/wiki/Q739302","display_name":"Production (economics)","level":2,"score":0.28949999809265137},{"id":"https://openalex.org/C31258907","wikidata":"https://www.wikidata.org/wiki/Q1301371","display_name":"Computer network","level":1,"score":0.2863999903202057},{"id":"https://openalex.org/C189783530","wikidata":"https://www.wikidata.org/wiki/Q352090","display_name":"CPU cache","level":3,"score":0.27619999647140503},{"id":"https://openalex.org/C38556500","wikidata":"https://www.wikidata.org/wiki/Q13404475","display_name":"Cache algorithms","level":4,"score":0.2648000121116638},{"id":"https://openalex.org/C120314980","wikidata":"https://www.wikidata.org/wiki/Q180634","display_name":"Distributed computing","level":1,"score":0.25699999928474426},{"id":"https://openalex.org/C2778565505","wikidata":"https://www.wikidata.org/wiki/Q2207566","display_name":"Spec#","level":2,"score":0.2515000104904175}],"mesh":[],"locations_count":2,"locations":[{"id":"pmh:oai:arXiv.org:2609.28870","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2609.28870","pdf_url":"https://arxiv.org/pdf/2609.28870","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,"listed_in":[],"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":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},{"id":"doi:10.48550/arxiv.2609.28870","doi":"https://doi.org/10.48550/arxiv.2609.28870","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2609.28870","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,"listed_in":[],"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"}],"best_oa_location":{"id":"pmh:oai:arXiv.org:2609.28870","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2609.28870","pdf_url":"https://arxiv.org/pdf/2609.28870","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,"listed_in":[],"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":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},"sustainable_development_goals":[],"study_designs":[],"awards":[],"funders":[],"has_content":{"grobid_xml":true,"pdf":true},"content_urls":{"pdf":"https://content.openalex.org/works/W7214397909.pdf","grobid_xml":"https://content.openalex.org/works/W7214397909.grobid-xml"},"referenced_works_count":4,"referenced_works":["https://openalex.org/W3001279689","https://openalex.org/W4378473852","https://openalex.org/W7077869950","https://openalex.org/W7124430182"],"related_works":["https://openalex.org/W7162785577","https://openalex.org/W7161788698","https://openalex.org/W7214748127","https://openalex.org/W7214734835","https://openalex.org/W7213607715","https://openalex.org/W7172399876","https://openalex.org/W7170060561","https://openalex.org/W7214147113","https://openalex.org/W7160657682","https://openalex.org/W4405033074"],"abstract_inverted_index":{"Long-running":[0],"LLM":[1],"applications":[2],"repeatedly":[3],"send":[4],"growing":[5],"context,":[6],"making":[7,74],"prefix":[8,63],"caching":[9,79],"critical":[10],"for":[11,51,131,137],"reducing":[12],"prefill":[13],"cost.":[14],"Yet":[15],"prefix-cache":[16,118],"behavior":[17],"under":[18],"agentic":[19],"workloads":[20],"remains":[21],"poorly":[22],"understood.":[23],"We":[24,100,144],"study":[25],"production":[26],"traces":[27,148],"from":[28],"two":[29,106],"companies":[30],"and":[31,38,88,105,140,149],"evaluate":[32],"14":[33],"eviction":[34,136,142],"algorithms":[35],"across":[36],"HBM-constrained":[37],"large":[39,44],"memory-pool":[40],"settings.":[41],"Despite":[42],"a":[43],"gap":[45],"to":[46,109,151],"Belady,":[47],"sophisticated":[48],"policies":[49],"designed":[50],"traditional":[52],"caches":[53],"provide":[54],"little":[55],"benefit":[56],"over":[57],"LRU.":[58],"The":[59],"reason":[60],"is":[61,65],"structural:":[62],"reuse":[64],"dominated":[66],"by":[67],"the":[68,102,147],"regular":[69],"pacing":[70],"of":[71],"active":[72],"sessions,":[73],"recency":[75,122],"unusually":[76],"predictive.":[77],"Prefix":[78],"nevertheless":[80],"introduces":[81],"new":[82],"challenges,":[83],"including":[84],"heavy-tailed":[85],"session":[86],"footprints":[87],"highly":[89],"variable":[90],"miss":[91],"costs":[92],"as":[93,123],"attention":[94],"computation":[95],"grows":[96],"with":[97],"sequence":[98],"length.":[99],"introduce":[101],"compute-savings":[103],"ratio":[104],"offline":[107],"oracles":[108],"quantify":[110],"these":[111],"effects.":[112],"Our":[113],"results":[114],"show":[115],"that":[116],"effective":[117],"management":[119],"should":[120],"retain":[121],"its":[124],"foundation":[125],"while":[126],"selectively":[127],"adding":[128],"quick":[129],"demotion":[130],"one-hit":[132],"prefixes,":[133],"compute-aware":[134],"partial":[135],"expensive":[138],"misses,":[139],"capacity-dependent":[141],"granularity.":[143],"will":[145],"release":[146],"simulator":[150],"support":[152],"future":[153],"research.":[154]},"counts_by_year":[{"year":2026,"cited_by_count":3}],"updated_date":"2026-11-07T11:25:07.637180","created_date":"2026-09-26T00:00:00"}
