In prefill-decode (PD) disaggregated LLM serving, each request is assigned to a decode worker after prefill. Existing decode routers balance only load; for mixture-of-experts (MoE) models this is incomplete: equally loaded workers can differ in latency, since each decode step loads the weights of every distinct expert its batch activates. We present ELDR, an expert-locality-aware decode router for PD-disaggregated MoE serving. From a request’s prefill expert activations, ELDR builds an expert signature predicting the experts it will activate during generation. Offline, balanced K-means partitions signature space across decode workers; online, locality-band routing sends each request to the least-loaded worker among those best matching its signature. A signature cache, co-indexed with the KV cache at KV-block granularity, keeps signatures exact under prefix caching. Implemented in vLLM and evaluated on deployments of up to 40 GPUs, ELDR reduces median TPOT by 5.9-13.9% over the strongest of four load-balancing baselines across three MoE models and two workloads, with model outputs unchanged.<\/p>\n","protected":false},"excerpt":{"rendered":"
In prefill-decode (PD) disaggregated LLM serving, each request is assigned to a decode worker after prefill. Existing decode routers balance only load; for mixture-of-experts (MoE) models this is incomplete: equally loaded workers can differ in latency, since each decode step loads the weights of every distinct expert its batch activates. We present ELDR, an expert-locality-aware […]<\/p>\n","protected":false},"featured_media":0,"template":"","meta":{"msr-url-field":"","msr-podcast-episode":"","msrModifiedDate":"","msrModifiedDateEnabled":false,"ep_exclude_from_search":false,"_classifai_error":"","msr-author-ordering":[{"type":"text","value":"Sang-Jun Choi","user_id":0},{"type":"text","value":"Sukmin Cho","user_id":0},{"type":"user_nicename","value":"Yifan Xiong","user_id":"40702"},{"type":"user_nicename","value":"Ziyue Yang","user_id":"41653"},{"type":"text","value":"Youngjin Kwon","user_id":0},{"type":"user_nicename","value":"Peng Cheng","user_id":"33225"}],"msr_publishername":"","msr_publisher_other":"","msr_booktitle":"","msr_chapter":"","msr_edition":"","msr_editors":"","msr_how_published":"arXiv","msr_isbn":"","msr_issue":"","msr_journal":"","msr_number":"","msr_organization":"","msr_pages_string":"","msr_page_range_start":"","msr_page_range_end":"","msr_series":"","msr_volume":"","msr_copyright":"","msr_conference_name":"","msr_doi":"","msr_arxiv_id":"2607.00466","msr_mag_id":"","msr_other_authors":"","msr_other_contributors":"","msr_speaker":"","msr_award":"","msr_affiliation":"","msr_institution":"","msr_host":"","msr_version":"","msr_duration":"","msr_release_tracker_id":"","msr_highlight_type":"","msr_date_display_format":"","msr_main_download_label":"","msr_external_link_label":"","msr_doi_label":"","msr_published_date":"2026-07-01","msr_startdate":"","msr_presentation_date":"","msr_highlight_text":"","msr_notes":"","msr_longbiography":"","msr_publicationurl":"","msr_external_url":"","msr_secondary_video_url":"","msr_conference_url":"","msr_journal_url":"","msr_year":2026,"msr_month":7,"msr_day":1,"msr_microsoftintellectualproperty":false,"msr_pub_id":"c02065f73957e41971cbf2da4a118fb93dd17c41","msr_publication_uploader":[{"type":"url","title":"https:\/\/arxiv.org\/abs\/2607.00466","label_id":243109,"id":false,"viewUrl":false}],"msr_related_uploader":[],"msr_original_fields_of_study":[],"msr_s2_paper_id":"c02065f73957e41971cbf2da4a118fb93dd17c41","msr_s2_pdf_url":"","msr_citation_count_updated":"","msr_citation_count":0,"msr_influential_citations":0,"msr_reference_count":42,"msr_s2_open_access":false,"msr_s2_author_ids":[],"msr_pub_ids":[{"provider":"s2","id":"c02065f73957e41971cbf2da4a118fb93dd17c41"},{"provider":"arxiv","id":"2607.00466"},{"provider":"corpusid","id":"289707129"}],"msr_hide_image_in_river":0,"footnotes":""},"msr-research-highlight":[],"research-area":[13560],"msr-publication-type":[270373],"msr-publisher":[],"msr-publication-cta":[],"msr-focus-area":[],"msr-locale":[268875],"msr-post-option":[],"msr-field-of-study":[246691,246793],"msr-conference":[],"msr-journal":[],"msr-impact-theme":[],"msr-pillar":[],"class_list":["post-1178888","msr-research-item","type-msr-research-item","status-publish","hentry","msr-research-area-programming-languages-software-engineering","msr-locale-en_us","msr-field-of-study-computer-science","msr-field-of-study-distributed-computing"],"msr_publishername":"","msr_edition":"","msr_affiliation":"","msr_published_date":"2026-07-01","msr_host":"","msr_duration":"","msr_version":"","msr_speaker":"","msr_other_contributors":"","msr_booktitle":"","msr_pages_string":"","msr_chapter":"","msr_isbn":"","msr_journal":"","msr_volume":"","msr_number":"","msr_editors":"","msr_series":"","msr_issue":"","msr_organization":"","msr_how_published":"arXiv","msr_notes":"","msr_highlight_text":"","msr_release_tracker_id":"","msr_original_fields_of_study":"","msr_download_urls":"","msr_external_url":"","msr_secondary_video_url":"","msr_longbiography":"","msr_microsoftintellectualproperty":0,"msr_main_download":"","msr_publicationurl":"","msr_doi":"","msr_publication_uploader":[{"type":"url","title":"https:\/\/arxiv.org\/abs\/2607.00466","label_id":243109,"id":false,"viewUrl":false}],"msr_related_uploader":[],"msr_citation_count":0,"msr_citation_count_updated":"","msr_s2_paper_id":"c02065f73957e41971cbf2da4a118fb93dd17c41","msr_influential_citations":0,"msr_reference_count":42,"msr_arxiv_id":"2607.00466","msr_s2_author_ids":[],"msr_s2_open_access":false,"msr_s2_pdf_url":null,"msr_attachments":[],"msr-author-ordering":[{"type":"text","value":"Sang-Jun Choi","user_id":0,"rest_url":false},{"type":"text","value":"Sukmin Cho","user_id":0,"rest_url":false},{"type":"user_nicename","value":"Yifan Xiong","user_id":40702,"rest_url":"https:\/\/find.codeghost.online\/en-us\/research\/wp-json\/microsoft-research\/v1\/researchers?person=Yifan Xiong"},{"type":"user_nicename","value":"Ziyue Yang","user_id":41653,"rest_url":"https:\/\/find.codeghost.online\/en-us\/research\/wp-json\/microsoft-research\/v1\/researchers?person=Ziyue Yang"},{"type":"text","value":"Youngjin Kwon","user_id":0,"rest_url":false},{"type":"user_nicename","value":"Peng Cheng","user_id":33225,"rest_url":"https:\/\/find.codeghost.online\/en-us\/research\/wp-json\/microsoft-research\/v1\/researchers?person=Peng Cheng"}],"msr_impact_theme":[],"msr_research_lab":[199560,199565],"msr_event":[],"msr_group":[],"msr_project":[],"publication":[],"video":[],"msr-tool":[],"msr_publication_type":"misc","related_content":[],"_links":{"self":[{"href":"https:\/\/find.codeghost.online\/en-us\/research\/wp-json\/wp\/v2\/msr-research-item\/1178888","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/find.codeghost.online\/en-us\/research\/wp-json\/wp\/v2\/msr-research-item"}],"about":[{"href":"https:\/\/find.codeghost.online\/en-us\/research\/wp-json\/wp\/v2\/types\/msr-research-item"}],"version-history":[{"count":2,"href":"https:\/\/find.codeghost.online\/en-us\/research\/wp-json\/wp\/v2\/msr-research-item\/1178888\/revisions"}],"predecessor-version":[{"id":1179048,"href":"https:\/\/find.codeghost.online\/en-us\/research\/wp-json\/wp\/v2\/msr-research-item\/1178888\/revisions\/1179048"}],"wp:attachment":[{"href":"https:\/\/find.codeghost.online\/en-us\/research\/wp-json\/wp\/v2\/media?parent=1178888"}],"wp:term":[{"taxonomy":"msr-research-highlight","embeddable":true,"href":"https:\/\/find.codeghost.online\/en-us\/research\/wp-json\/wp\/v2\/msr-research-highlight?post=1178888"},{"taxonomy":"msr-research-area","embeddable":true,"href":"https:\/\/find.codeghost.online\/en-us\/research\/wp-json\/wp\/v2\/research-area?post=1178888"},{"taxonomy":"msr-publication-type","embeddable":true,"href":"https:\/\/find.codeghost.online\/en-us\/research\/wp-json\/wp\/v2\/msr-publication-type?post=1178888"},{"taxonomy":"msr-publisher","embeddable":true,"href":"https:\/\/find.codeghost.online\/en-us\/research\/wp-json\/wp\/v2\/msr-publisher?post=1178888"},{"taxonomy":"msr-publication-cta","embeddable":true,"href":"https:\/\/find.codeghost.online\/en-us\/research\/wp-json\/wp\/v2\/msr-publication-cta?post=1178888"},{"taxonomy":"msr-focus-area","embeddable":true,"href":"https:\/\/find.codeghost.online\/en-us\/research\/wp-json\/wp\/v2\/msr-focus-area?post=1178888"},{"taxonomy":"msr-locale","embeddable":true,"href":"https:\/\/find.codeghost.online\/en-us\/research\/wp-json\/wp\/v2\/msr-locale?post=1178888"},{"taxonomy":"msr-post-option","embeddable":true,"href":"https:\/\/find.codeghost.online\/en-us\/research\/wp-json\/wp\/v2\/msr-post-option?post=1178888"},{"taxonomy":"msr-field-of-study","embeddable":true,"href":"https:\/\/find.codeghost.online\/en-us\/research\/wp-json\/wp\/v2\/msr-field-of-study?post=1178888"},{"taxonomy":"msr-conference","embeddable":true,"href":"https:\/\/find.codeghost.online\/en-us\/research\/wp-json\/wp\/v2\/msr-conference?post=1178888"},{"taxonomy":"msr-journal","embeddable":true,"href":"https:\/\/find.codeghost.online\/en-us\/research\/wp-json\/wp\/v2\/msr-journal?post=1178888"},{"taxonomy":"msr-impact-theme","embeddable":true,"href":"https:\/\/find.codeghost.online\/en-us\/research\/wp-json\/wp\/v2\/msr-impact-theme?post=1178888"},{"taxonomy":"msr-pillar","embeddable":true,"href":"https:\/\/find.codeghost.online\/en-us\/research\/wp-json\/wp\/v2\/msr-pillar?post=1178888"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}