{"id":"https://openalex.org/W7160606018","doi":"https://doi.org/10.48550/arxiv.2605.06055","title":"Relay Buffer Independent Communication over Pooled HBM for Efficient MoE Inference on Ascend","display_name":"Relay Buffer Independent Communication over Pooled HBM for Efficient MoE Inference on Ascend","publication_year":2026,"publication_date":"2026-05-07","ids":{"openalex":"https://openalex.org/W7160606018","doi":"https://doi.org/10.48550/arxiv.2605.06055"},"language":null,"primary_location":{"id":"doi:10.48550/arxiv.2605.06055","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.06055","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.2605.06055","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5042625562","display_name":"Tianlun Hu","orcid":"https://orcid.org/0000-0001-9303-0089"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Hu, Tianlun","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5135664076","display_name":"Tiancheng Hu","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Hu, Tiancheng","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5135698515","display_name":"Shengsheng Litang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Litang, Shengsheng","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5135649332","display_name":"Sheng Wang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Wang, Sheng","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5111600144","display_name":"Xiaoming Bao","orcid":"https://orcid.org/0000-0003-1486-8305"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Bao, Xiaoming","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5135724611","display_name":"Yuxing Li","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Li, Yuxing","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5135665046","display_name":"Wei Wang","orcid":"https://orcid.org/0009-0001-3925-3924"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Wang, Wei","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5075481740","display_name":"Zhongzhe Hu","orcid":"https://orcid.org/0000-0002-6708-3942"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Hu, Zhongzhe","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5135714106","display_name":"Lijun Li","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Li, Lijun","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5038216935","display_name":"Hongwei Sun","orcid":"https://orcid.org/0000-0001-9116-3865"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Sun, Hongwei","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5135642122","display_name":"Jingbin Zhou\\\\","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zhou, Jingbin","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/T11896","display_name":"Opportunistic and Delay-Tolerant Networks","score":0.1696999967098236,"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"}},"topics":[{"id":"https://openalex.org/T11896","display_name":"Opportunistic and Delay-Tolerant Networks","score":0.1696999967098236,"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/T10772","display_name":"Distributed systems and fault tolerance","score":0.15119999647140503,"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/T10764","display_name":"Privacy-Preserving Technologies in Data","score":0.07750000059604645,"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/security-token","display_name":"Security token","score":0.756600022315979},{"id":"https://openalex.org/keywords/relay","display_name":"Relay","score":0.6777999997138977},{"id":"https://openalex.org/keywords/token-passing","display_name":"Token passing","score":0.6227999925613403},{"id":"https://openalex.org/keywords/scheduling","display_name":"Scheduling (production processes)","score":0.6047999858856201},{"id":"https://openalex.org/keywords/latency","display_name":"Latency (audio)","score":0.5767999887466431},{"id":"https://openalex.org/keywords/schedule","display_name":"Schedule","score":0.5016999840736389},{"id":"https://openalex.org/keywords/inference","display_name":"Inference","score":0.4945000112056732}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.8172000050544739},{"id":"https://openalex.org/C48145219","wikidata":"https://www.wikidata.org/wiki/Q1335365","display_name":"Security token","level":2,"score":0.756600022315979},{"id":"https://openalex.org/C2778156585","wikidata":"https://www.wikidata.org/wiki/Q174053","display_name":"Relay","level":3,"score":0.6777999997138977},{"id":"https://openalex.org/C115067241","wikidata":"https://www.wikidata.org/wiki/Q1639854","display_name":"Token passing","level":3,"score":0.6227999925613403},{"id":"https://openalex.org/C206729178","wikidata":"https://www.wikidata.org/wiki/Q2271896","display_name":"Scheduling (production processes)","level":2,"score":0.6047999858856201},{"id":"https://openalex.org/C82876162","wikidata":"https://www.wikidata.org/wiki/Q17096504","display_name":"Latency (audio)","level":2,"score":0.5767999887466431},{"id":"https://openalex.org/C68387754","wikidata":"https://www.wikidata.org/wiki/Q7271585","display_name":"Schedule","level":2,"score":0.5016999840736389},{"id":"https://openalex.org/C2776214188","wikidata":"https://www.wikidata.org/wiki/Q408386","display_name":"Inference","level":2,"score":0.4945000112056732},{"id":"https://openalex.org/C79403827","wikidata":"https://www.wikidata.org/wiki/Q3988","display_name":"Real-time computing","level":1,"score":0.4113999903202057},{"id":"https://openalex.org/C48103436","wikidata":"https://www.wikidata.org/wiki/Q599031","display_name":"State (computer science)","level":2,"score":0.38929998874664307},{"id":"https://openalex.org/C120314980","wikidata":"https://www.wikidata.org/wiki/Q180634","display_name":"Distributed computing","level":1,"score":0.3882000148296356},{"id":"https://openalex.org/C31258907","wikidata":"https://www.wikidata.org/wiki/Q1301371","display_name":"Computer network","level":1,"score":0.376800000667572},{"id":"https://openalex.org/C2778562939","wikidata":"https://www.wikidata.org/wiki/Q1298791","display_name":"Synchronization (alternating current)","level":3,"score":0.36090001463890076},{"id":"https://openalex.org/C95203288","wikidata":"https://www.wikidata.org/wiki/Q221682","display_name":"Semaphore","level":2,"score":0.32330000400543213},{"id":"https://openalex.org/C102462485","wikidata":"https://www.wikidata.org/wiki/Q904595","display_name":"Token bus network","level":3,"score":0.28060001134872437},{"id":"https://openalex.org/C46743427","wikidata":"https://www.wikidata.org/wiki/Q1341685","display_name":"Inference engine","level":3,"score":0.27970001101493835},{"id":"https://openalex.org/C2777904410","wikidata":"https://www.wikidata.org/wiki/Q7397","display_name":"Software","level":2,"score":0.27410000562667847},{"id":"https://openalex.org/C72434380","wikidata":"https://www.wikidata.org/wiki/Q230930","display_name":"State space","level":2,"score":0.2720000147819519},{"id":"https://openalex.org/C192126672","wikidata":"https://www.wikidata.org/wiki/Q1068715","display_name":"Telecommunications network","level":2,"score":0.2703999876976013}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2605.06055","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.06055","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.2605.06055","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.06055","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":[],"awards":[],"funders":[],"has_content":{"grobid_xml":false,"pdf":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"Mixture-of-Experts":[0],"(MoE)":[1],"inference":[2,61],"requires":[3],"large-scale":[4],"token":[5,171,178],"exchange":[6],"across":[7],"devices,":[8],"making":[9],"dispatch":[10,69,154],"and":[11,18,28,46,70,79,92,100,112,139,155,180,204],"combine":[12,71,156],"major":[13],"bottlenecks":[14],"in":[15,158],"both":[16,159],"prefill":[17,130],"decode.":[19],"Beyond":[20],"network":[21],"transfer,":[22],"routing-driven":[23],"layout":[24],"transformation,":[25],"temporary":[26],"relay,":[27],"output":[29,177,205],"restoration":[30,206],"can":[31],"add":[32],"substantial":[33],"overhead.":[34],"Existing":[35],"MoE":[36,60,127,150,216],"communication":[37,57],"paths":[38],"are":[39],"often":[40],"buffer-centric,":[41],"using":[42],"explicit":[43],"inter-process":[44],"relay":[45,99],"reordering":[47,101],"buffers":[48,102],"around":[49,72,207],"collective":[50],"transfer.":[51],"This":[52],"report":[53],"presents":[54],"a":[55,129,140],"relay-buffer-free":[56],"design":[58,67,118],"for":[59,122,136,144,214],"acceleration":[62],"on":[63,87,148,194],"Ascend":[64],"systems.":[65],"The":[66],"reorganizes":[68],"direct":[73,80],"placement":[74],"into":[75],"destination":[76],"expert":[77,84,208],"windows":[78],"reading":[81],"from":[82],"remote":[83],"windows.":[85],"Built":[86],"globally":[88,197],"pooled":[89],"high-bandwidth":[90],"memory":[91],"symmetric-memory":[93],"allocation,":[94],"it":[95],"removes":[96],"most":[97],"intermediate":[98,202],"while":[103],"retaining":[104],"only":[105],"lightweight":[106],"control":[107],"state,":[108],"including":[109],"counts,":[110],"offsets,":[111],"synchronization":[113],"metadata.":[114],"We":[115],"instantiate":[116],"the":[117,123,162,165,182],"as":[119],"two":[120],"schedules":[121],"main":[124],"phases":[125],"of":[126],"inference:":[128],"schedule":[131,143],"with":[132,196],"richer":[133],"planning":[134],"state":[135],"throughput-oriented":[137],"execution,":[138],"compact":[141],"decode":[142],"latency-sensitive":[145],"execution.":[146],"Experiments":[147],"Ascend-based":[149],"workloads":[151],"show":[152],"reduced":[153],"latency":[157,188],"settings.":[160],"At":[161],"serving":[163],"level,":[164],"implementation":[166],"improves":[167],"time":[168,175],"to":[169],"first":[170],"(TTFT),":[172],"preserves":[173],"competitive":[174],"per":[176],"(TPOT),":[179],"enlarges":[181],"feasible":[183],"scheduling":[184],"space":[185],"under":[186],"practical":[187],"constraints.":[189],"These":[190],"results":[191],"indicate":[192],"that,":[193],"platforms":[195],"addressable":[198],"device":[199],"memory,":[200],"reducing":[201],"buffering":[203],"execution":[209],"is":[210],"an":[211],"effective":[212],"direction":[213],"accelerating":[215],"inference.":[217]},"counts_by_year":[],"updated_date":"2026-07-01T06:00:48.157686","created_date":"2026-05-09T00:00:00"}