{"id":"https://openalex.org/W4404400716","doi":"https://doi.org/10.1145/3694715.3695963","title":"Apparate: Rethinking Early Exits to Tame Latency-Throughput Tensions in ML Serving","display_name":"Apparate: Rethinking Early Exits to Tame Latency-Throughput Tensions in ML Serving","publication_year":2024,"publication_date":"2024-11-04","ids":{"openalex":"https://openalex.org/W4404400716","doi":"https://doi.org/10.1145/3694715.3695963"},"language":"en","primary_location":{"id":"doi:10.1145/3694715.3695963","is_oa":true,"landing_page_url":"https://doi.org/10.1145/3694715.3695963","pdf_url":"https://dl.acm.org/doi/pdf/10.1145/3694715.3695963","source":null,"license":"cc-by-sa","license_id":"https://openalex.org/licenses/cc-by-sa","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Proceedings of the ACM SIGOPS 30th Symposium on Operating Systems Principles","raw_type":"proceedings-article"},"type":"article","indexed_in":["crossref"],"open_access":{"is_oa":true,"oa_status":"gold","oa_url":"https://dl.acm.org/doi/pdf/10.1145/3694715.3695963","any_repository_has_fulltext":null},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5103107119","display_name":"Yinwei Dai","orcid":"https://orcid.org/0000-0002-9291-2060"},"institutions":[{"id":"https://openalex.org/I20089843","display_name":"Princeton University","ror":"https://ror.org/00hx57361","country_code":"US","type":"education","lineage":["https://openalex.org/I20089843"]}],"countries":["US"],"is_corresponding":true,"raw_author_name":"Yinwei Dai","raw_affiliation_strings":["Princeton University, Princeton, New Jersey, United States"],"affiliations":[{"raw_affiliation_string":"Princeton University, Princeton, New Jersey, United States","institution_ids":["https://openalex.org/I20089843"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5100638027","display_name":"Rui Pan","orcid":"https://orcid.org/0000-0002-6973-3259"},"institutions":[{"id":"https://openalex.org/I20089843","display_name":"Princeton University","ror":"https://ror.org/00hx57361","country_code":"US","type":"education","lineage":["https://openalex.org/I20089843"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Rui Pan","raw_affiliation_strings":["Princeton University, Princeton, New Jersey, United States"],"affiliations":[{"raw_affiliation_string":"Princeton University, Princeton, New Jersey, United States","institution_ids":["https://openalex.org/I20089843"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5090733623","display_name":"Anand Iyer","orcid":"https://orcid.org/0009-0009-3700-2994"},"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"]},{"id":"https://openalex.org/I2800444561","display_name":"Atlanta Technical College","ror":"https://ror.org/01s3vfp47","country_code":"US","type":"education","lineage":["https://openalex.org/I2800444561"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Anand Iyer","raw_affiliation_strings":["Georgia Tech, Atlanta, Georgia, United States"],"affiliations":[{"raw_affiliation_string":"Georgia Tech, Atlanta, Georgia, United States","institution_ids":["https://openalex.org/I2800444561","https://openalex.org/I130701444"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5100399870","display_name":"Kai Li","orcid":"https://orcid.org/0000-0002-2095-7024"},"institutions":[{"id":"https://openalex.org/I20089843","display_name":"Princeton University","ror":"https://ror.org/00hx57361","country_code":"US","type":"education","lineage":["https://openalex.org/I20089843"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Kai Li","raw_affiliation_strings":["Princeton University, Princeton, New Jersey, USA"],"affiliations":[{"raw_affiliation_string":"Princeton University, Princeton, New Jersey, USA","institution_ids":["https://openalex.org/I20089843"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5053593890","display_name":"Ravi Netravali","orcid":"https://orcid.org/0000-0001-7002-5033"},"institutions":[{"id":"https://openalex.org/I20089843","display_name":"Princeton University","ror":"https://ror.org/00hx57361","country_code":"US","type":"education","lineage":["https://openalex.org/I20089843"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Ravi Netravali","raw_affiliation_strings":["Princeton University, Princeton, New Jersey, United States"],"affiliations":[{"raw_affiliation_string":"Princeton University, Princeton, New Jersey, United States","institution_ids":["https://openalex.org/I20089843"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":5,"corresponding_author_ids":["https://openalex.org/A5103107119"],"corresponding_institution_ids":["https://openalex.org/I20089843"],"apc_list":null,"apc_paid":null,"fwci":2.0153,"has_fulltext":false,"cited_by_count":4,"citation_normalized_percentile":{"value":0.88642037,"is_in_top_1_percent":false,"is_in_top_10_percent":false},"cited_by_percentile_year":{"min":97,"max":98},"biblio":{"volume":null,"issue":null,"first_page":"607","last_page":"623"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T10054","display_name":"Parallel Computing and Optimization Techniques","score":0.9984999895095825,"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":0.9984999895095825,"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/T10743","display_name":"Software Testing and Debugging Techniques","score":0.998199999332428,"subfield":{"id":"https://openalex.org/subfields/1712","display_name":"Software"},"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/T10028","display_name":"Topic Modeling","score":0.9976999759674072,"subfield":{"id":"https://openalex.org/subfields/1702","display_name":"Artificial Intelligence"},"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/throughput","display_name":"Throughput","score":0.7119273543357849},{"id":"https://openalex.org/keywords/latency","display_name":"Latency (audio)","score":0.7089996337890625},{"id":"https://openalex.org/keywords/computer-science","display_name":"Computer science","score":0.6727238297462463},{"id":"https://openalex.org/keywords/computer-network","display_name":"Computer network","score":0.3744814991950989},{"id":"https://openalex.org/keywords/operating-system","display_name":"Operating system","score":0.20940178632736206},{"id":"https://openalex.org/keywords/telecommunications","display_name":"Telecommunications","score":0.1834034025669098},{"id":"https://openalex.org/keywords/wireless","display_name":"Wireless","score":0.07272252440452576}],"concepts":[{"id":"https://openalex.org/C157764524","wikidata":"https://www.wikidata.org/wiki/Q1383412","display_name":"Throughput","level":3,"score":0.7119273543357849},{"id":"https://openalex.org/C82876162","wikidata":"https://www.wikidata.org/wiki/Q17096504","display_name":"Latency (audio)","level":2,"score":0.7089996337890625},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.6727238297462463},{"id":"https://openalex.org/C31258907","wikidata":"https://www.wikidata.org/wiki/Q1301371","display_name":"Computer network","level":1,"score":0.3744814991950989},{"id":"https://openalex.org/C111919701","wikidata":"https://www.wikidata.org/wiki/Q9135","display_name":"Operating system","level":1,"score":0.20940178632736206},{"id":"https://openalex.org/C76155785","wikidata":"https://www.wikidata.org/wiki/Q418","display_name":"Telecommunications","level":1,"score":0.1834034025669098},{"id":"https://openalex.org/C555944384","wikidata":"https://www.wikidata.org/wiki/Q249","display_name":"Wireless","level":2,"score":0.07272252440452576}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.1145/3694715.3695963","is_oa":true,"landing_page_url":"https://doi.org/10.1145/3694715.3695963","pdf_url":"https://dl.acm.org/doi/pdf/10.1145/3694715.3695963","source":null,"license":"cc-by-sa","license_id":"https://openalex.org/licenses/cc-by-sa","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Proceedings of the ACM SIGOPS 30th Symposium on Operating Systems Principles","raw_type":"proceedings-article"}],"best_oa_location":{"id":"doi:10.1145/3694715.3695963","is_oa":true,"landing_page_url":"https://doi.org/10.1145/3694715.3695963","pdf_url":"https://dl.acm.org/doi/pdf/10.1145/3694715.3695963","source":null,"license":"cc-by-sa","license_id":"https://openalex.org/licenses/cc-by-sa","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Proceedings of the ACM SIGOPS 30th Symposium on Operating Systems Principles","raw_type":"proceedings-article"},"sustainable_development_goals":[],"awards":[],"funders":[],"has_content":{"grobid_xml":false,"pdf":true},"content_urls":{"pdf":"https://content.openalex.org/works/W4404400716.pdf"},"referenced_works_count":32,"referenced_works":["https://openalex.org/W95608104","https://openalex.org/W2160815625","https://openalex.org/W2734941459","https://openalex.org/W2772948367","https://openalex.org/W2914209329","https://openalex.org/W2931743911","https://openalex.org/W2962677625","https://openalex.org/W2969950143","https://openalex.org/W2981852735","https://openalex.org/W2982157693","https://openalex.org/W3008591352","https://openalex.org/W3034292689","https://openalex.org/W3035030897","https://openalex.org/W3035038672","https://openalex.org/W3081751010","https://openalex.org/W3095488153","https://openalex.org/W3154971029","https://openalex.org/W3158444059","https://openalex.org/W3160106041","https://openalex.org/W4212774754","https://openalex.org/W4240862739","https://openalex.org/W4285595056","https://openalex.org/W4288089799","https://openalex.org/W4288348249","https://openalex.org/W4387321091","https://openalex.org/W4388874804","https://openalex.org/W4389523983","https://openalex.org/W4402671987","https://openalex.org/W4404400640","https://openalex.org/W6762481227","https://openalex.org/W6840061620","https://openalex.org/W6888476208"],"related_works":["https://openalex.org/W4391375266","https://openalex.org/W2899084033","https://openalex.org/W2748952813","https://openalex.org/W2390279801","https://openalex.org/W4391913857","https://openalex.org/W2358668433","https://openalex.org/W4396701345","https://openalex.org/W2376932109","https://openalex.org/W2001405890","https://openalex.org/W4396696052"],"abstract_inverted_index":{"Machine":[0],"learning":[1],"(ML)":[2],"inference":[3,69],"platforms":[4],"are":[5],"tasked":[6],"with":[7,93,100],"balancing":[8],"two":[9],"competing":[10],"goals:":[11],"ensuring":[12],"high":[13],"throughput":[14],"given":[15],"many":[16],"requests,":[17],"and":[18,39,80,104,123,133,138,142],"delivering":[19],"low-latency":[20],"responses":[21],"to":[22,34,44,59,113],"support":[23],"interactive":[24],"applications.":[25],"Unfortunately,":[26],"existing":[27],"platform":[28],"knobs":[29],"(e.g.,":[30],"batch":[31],"sizes)":[32],"fail":[33],"ease":[35],"this":[36],"fundamental":[37],"tension,":[38],"instead":[40],"only":[41],"enable":[42],"users":[43],"harshly":[45],"trade":[46],"off":[47],"one":[48],"property":[49],"for":[50,135,148],"the":[51,65,101],"other.":[52],"This":[53],"paper":[54],"explores":[55],"an":[56],"alternate":[57],"strategy":[58],"taming":[60],"throughput-latency":[61],"tradeoffs":[62],"by":[63,131,146],"changing":[64],"granularity":[66],"at":[67,95],"which":[68],"is":[70],"performed.":[71],"We":[72],"present":[73],"Apparate,":[74],"a":[75],"system":[76],"that":[77,107,117],"automatically":[78],"applies":[79],"manages":[81],"early":[82],"exits":[83,112],"(EEs)":[84],"in":[85],"ML":[86],"models,":[87],"whereby":[88],"certain":[89],"inputs":[90],"can":[91],"exit":[92],"results":[94],"intermediate":[96],"layers.":[97],"To":[98],"cope":[99],"time-varying":[102],"overhead":[103],"accuracy":[105,157],"challenges":[106],"EEs":[108],"bring,":[109],"Apparate":[110,126],"repurposes":[111],"provide":[114],"continual":[115],"feedback":[116],"powers":[118],"several":[119],"novel":[120],"runtime":[121],"monitoring":[122],"adaptation":[124],"strategies.":[125],"lowers":[127],"median":[128,143],"response":[129],"latencies":[130,145],"40.5--91.5%":[132],"10.0--24.2%":[134],"diverse":[136],"CV":[137],"NLP":[139],"classification":[140],"workloads,":[141],"time-per-token":[144],"22.6--77.9%":[147],"generative":[149],"scenarios,":[150],"without":[151],"affecting":[152],"throughputs":[153],"or":[154],"violating":[155],"tight":[156],"constraints.":[158]},"counts_by_year":[{"year":2025,"cited_by_count":4}],"updated_date":"2026-03-11T06:11:40.159057","created_date":"2024-11-15T00:00:00"}