{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:DQBXFOZBJBIHUJHG2XPZRM4G23","short_pith_number":"pith:DQBXFOZB","schema_version":"1.0","canonical_sha256":"1c0372bb2148507a24e6d5df98b386d6ccba74b79a73490c0e9a21cfeb42ee00","source":{"kind":"arxiv","id":"2504.09285","version":2},"attestation_state":"computed","paper":{"title":"DynaServe: Unified and Elastic Execution for Dynamic Disaggregated LLM Serving","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.DC","authors_text":"Chaoyi Ruan, Cheng Li, Dongqi Tian, Jialin Li, Yandong Shi, Yinhe Chen, Yongji Wu","submitted_at":"2025-04-12T17:09:54Z","abstract_excerpt":"LLM inference must meet strict latency SLOs (e.g., 100 ms P99 time-between-tokens) while maximizing goodput. Yet, real-world variability in prompt and response lengths skews compute-intensive prefill and memory-bound decode phases, making both colocated (even with chunked prefill) and disaggregated deployments unable to simultaneously deliver low tail latency and high throughput.\n  We introduce DynaServe, a high-performance LLM serving system built atop vLLM that unifies and extends both paradigms for maximizing goodput under SLO constraints, when handling unbalanced and dynamic workloads. It "},"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":"2504.09285","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.DC","submitted_at":"2025-04-12T17:09:54Z","cross_cats_sorted":[],"title_canon_sha256":"e5802331ecc6b2a56ca33105955ce1c2960a5db45f5f69e83560d745499da5e0","abstract_canon_sha256":"cdab29235b961504121057ab81c37f26a31a1fd9ed4f93151ddb19948f594f94"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:07:23.543169Z","signature_b64":"t+MmJsMpzY9q2Qc6viJueDcqqyD9ZwsatcJS5fXCt5E/COuxqqdvWDOpg8Rwr/j1YqObsuqmEcJGldHLT/eUBw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"1c0372bb2148507a24e6d5df98b386d6ccba74b79a73490c0e9a21cfeb42ee00","last_reissued_at":"2026-07-05T11:07:23.542680Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:07:23.542680Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"DynaServe: Unified and Elastic Execution for Dynamic Disaggregated LLM Serving","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.DC","authors_text":"Chaoyi Ruan, Cheng Li, Dongqi Tian, Jialin Li, Yandong Shi, Yinhe Chen, Yongji Wu","submitted_at":"2025-04-12T17:09:54Z","abstract_excerpt":"LLM inference must meet strict latency SLOs (e.g., 100 ms P99 time-between-tokens) while maximizing goodput. Yet, real-world variability in prompt and response lengths skews compute-intensive prefill and memory-bound decode phases, making both colocated (even with chunked prefill) and disaggregated deployments unable to simultaneously deliver low tail latency and high throughput.\n  We introduce DynaServe, a high-performance LLM serving system built atop vLLM that unifies and extends both paradigms for maximizing goodput under SLO constraints, when handling unbalanced and dynamic workloads. It "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2504.09285","kind":"arxiv","version":2},"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/2504.09285/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":"2504.09285","created_at":"2026-07-05T11:07:23.542736+00:00"},{"alias_kind":"arxiv_version","alias_value":"2504.09285v2","created_at":"2026-07-05T11:07:23.542736+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2504.09285","created_at":"2026-07-05T11:07:23.542736+00:00"},{"alias_kind":"pith_short_12","alias_value":"DQBXFOZBJBIH","created_at":"2026-07-05T11:07:23.542736+00:00"},{"alias_kind":"pith_short_16","alias_value":"DQBXFOZBJBIHUJHG","created_at":"2026-07-05T11:07:23.542736+00:00"},{"alias_kind":"pith_short_8","alias_value":"DQBXFOZB","created_at":"2026-07-05T11:07:23.542736+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":2,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2607.08565","citing_title":"SMetric: Rethink LLM Scheduling for Serving Agents with Balanced Session-centric Scheduling","ref_index":41,"is_internal_anchor":true},{"citing_arxiv_id":"2606.00735","citing_title":"ViBE: Co-Optimizing Workload Skew and Hardware Variability for MoE Serving","ref_index":38,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/DQBXFOZBJBIHUJHG2XPZRM4G23","json":"https://pith.science/pith/DQBXFOZBJBIHUJHG2XPZRM4G23.json","graph_json":"https://pith.science/api/pith-number/DQBXFOZBJBIHUJHG2XPZRM4G23/graph.json","events_json":"https://pith.science/api/pith-number/DQBXFOZBJBIHUJHG2XPZRM4G23/events.json","paper":"https://pith.science/paper/DQBXFOZB"},"agent_actions":{"view_html":"https://pith.science/pith/DQBXFOZBJBIHUJHG2XPZRM4G23","download_json":"https://pith.science/pith/DQBXFOZBJBIHUJHG2XPZRM4G23.json","view_paper":"https://pith.science/paper/DQBXFOZB","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2504.09285&json=true","fetch_graph":"https://pith.science/api/pith-number/DQBXFOZBJBIHUJHG2XPZRM4G23/graph.json","fetch_events":"https://pith.science/api/pith-number/DQBXFOZBJBIHUJHG2XPZRM4G23/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/DQBXFOZBJBIHUJHG2XPZRM4G23/action/timestamp_anchor","attest_storage":"https://pith.science/pith/DQBXFOZBJBIHUJHG2XPZRM4G23/action/storage_attestation","attest_author":"https://pith.science/pith/DQBXFOZBJBIHUJHG2XPZRM4G23/action/author_attestation","sign_citation":"https://pith.science/pith/DQBXFOZBJBIHUJHG2XPZRM4G23/action/citation_signature","submit_replication":"https://pith.science/pith/DQBXFOZBJBIHUJHG2XPZRM4G23/action/replication_record"}},"created_at":"2026-07-05T11:07:23.542736+00:00","updated_at":"2026-07-05T11:07:23.542736+00:00"}