{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:UV3OFQJQV4TGJ5OXGM7UTYA6TG","short_pith_number":"pith:UV3OFQJQ","schema_version":"1.0","canonical_sha256":"a576e2c130af2664f5d7333f49e01e99917a3ba5c01e493677068cfa14f5629a","source":{"kind":"arxiv","id":"2503.20552","version":1},"attestation_state":"computed","paper":{"title":"Injecting Adrenaline into LLM Serving: Boosting Resource Utilization and Throughput via Attention Disaggregation","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.DC","authors_text":"Pengfei Zuo, Xu Chen, Yunkai Liang, Zhangyu Chen, Zhi Zhou, Zhou Yu","submitted_at":"2025-03-26T13:48:35Z","abstract_excerpt":"In large language model (LLM) serving systems, executing each request consists of two phases: the compute-intensive prefill phase and the memory-intensive decoding phase. To prevent performance interference between the two phases, current LLM serving systems typically adopt prefill-decoding disaggregation, where the two phases are split across separate machines. However, we observe this approach leads to significant resource underutilization. Specifically, prefill instances that are compute-intensive suffer from low memory utilization, while decoding instances that are memory-intensive experie"},"verification_status":{"content_addressed":true,"pith_receipt":true,"author_attested":false,"weak_author_claims":0,"strong_author_claims":0,"externally_anchored":false,"storage_verified":false,"citation_signatures":0,"replication_records":0,"graph_snapshot":true,"references_resolved":false,"formal_links_present":false},"canonical_record":{"source":{"id":"2503.20552","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.DC","submitted_at":"2025-03-26T13:48:35Z","cross_cats_sorted":["cs.LG"],"title_canon_sha256":"5a5b4630353edc24338babb2bc7fb0b4b49d4bedcd503cfa4232c844612c58c7","abstract_canon_sha256":"ce3cd4e066f480773147fc8e67aa16ffe4e79113de2e3b3ab6d0a3d489a02cea"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T10:39:40.109755Z","signature_b64":"a8GrtiL5TJw7bSGxYYs2f+6NMPiRuwcs2dL+YadkixM9rDWxrXYiVkOFy3x5lL3fBDC3GM5O5NY0+WnXY4b8Cg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"a576e2c130af2664f5d7333f49e01e99917a3ba5c01e493677068cfa14f5629a","last_reissued_at":"2026-07-05T10:39:40.109263Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T10:39:40.109263Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Injecting Adrenaline into LLM Serving: Boosting Resource Utilization and Throughput via Attention Disaggregation","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.DC","authors_text":"Pengfei Zuo, Xu Chen, Yunkai Liang, Zhangyu Chen, Zhi Zhou, Zhou Yu","submitted_at":"2025-03-26T13:48:35Z","abstract_excerpt":"In large language model (LLM) serving systems, executing each request consists of two phases: the compute-intensive prefill phase and the memory-intensive decoding phase. To prevent performance interference between the two phases, current LLM serving systems typically adopt prefill-decoding disaggregation, where the two phases are split across separate machines. However, we observe this approach leads to significant resource underutilization. Specifically, prefill instances that are compute-intensive suffer from low memory utilization, while decoding instances that are memory-intensive experie"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2503.20552","kind":"arxiv","version":1},"verdict":{"id":null,"model_set":{},"created_at":null,"strongest_claim":"","one_line_summary":"","pipeline_version":null,"weakest_assumption":"","pith_extraction_headline":""},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2503.20552/integrity.json","findings":[],"available":true,"detectors_run":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938"},"references":{"count":0,"sample":[],"resolved_work":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","internal_anchors":0},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"},"aliases":[{"alias_kind":"arxiv","alias_value":"2503.20552","created_at":"2026-07-05T10:39:40.109334+00:00"},{"alias_kind":"arxiv_version","alias_value":"2503.20552v1","created_at":"2026-07-05T10:39:40.109334+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2503.20552","created_at":"2026-07-05T10:39:40.109334+00:00"},{"alias_kind":"pith_short_12","alias_value":"UV3OFQJQV4TG","created_at":"2026-07-05T10:39:40.109334+00:00"},{"alias_kind":"pith_short_16","alias_value":"UV3OFQJQV4TGJ5OX","created_at":"2026-07-05T10:39:40.109334+00:00"},{"alias_kind":"pith_short_8","alias_value":"UV3OFQJQ","created_at":"2026-07-05T10:39:40.109334+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.01502","citing_title":"Move the Query, Not the Cache: Characterizing Cross-Instance Latent Attention Redistribution Across GPU Fabrics","ref_index":21,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/UV3OFQJQV4TGJ5OXGM7UTYA6TG","json":"https://pith.science/pith/UV3OFQJQV4TGJ5OXGM7UTYA6TG.json","graph_json":"https://pith.science/api/pith-number/UV3OFQJQV4TGJ5OXGM7UTYA6TG/graph.json","events_json":"https://pith.science/api/pith-number/UV3OFQJQV4TGJ5OXGM7UTYA6TG/events.json","paper":"https://pith.science/paper/UV3OFQJQ"},"agent_actions":{"view_html":"https://pith.science/pith/UV3OFQJQV4TGJ5OXGM7UTYA6TG","download_json":"https://pith.science/pith/UV3OFQJQV4TGJ5OXGM7UTYA6TG.json","view_paper":"https://pith.science/paper/UV3OFQJQ","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2503.20552&json=true","fetch_graph":"https://pith.science/api/pith-number/UV3OFQJQV4TGJ5OXGM7UTYA6TG/graph.json","fetch_events":"https://pith.science/api/pith-number/UV3OFQJQV4TGJ5OXGM7UTYA6TG/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/UV3OFQJQV4TGJ5OXGM7UTYA6TG/action/timestamp_anchor","attest_storage":"https://pith.science/pith/UV3OFQJQV4TGJ5OXGM7UTYA6TG/action/storage_attestation","attest_author":"https://pith.science/pith/UV3OFQJQV4TGJ5OXGM7UTYA6TG/action/author_attestation","sign_citation":"https://pith.science/pith/UV3OFQJQV4TGJ5OXGM7UTYA6TG/action/citation_signature","submit_replication":"https://pith.science/pith/UV3OFQJQV4TGJ5OXGM7UTYA6TG/action/replication_record"}},"created_at":"2026-07-05T10:39:40.109334+00:00","updated_at":"2026-07-05T10:39:40.109334+00:00"}