{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:2345I7IWUWGVNWQU22MAFUBZUP","short_pith_number":"pith:2345I7IW","schema_version":"1.0","canonical_sha256":"d6f9d47d16a58d56da14d69802d039a3f6e65609fa4ff3e59ffdfb2a99d01155","source":{"kind":"arxiv","id":"2607.02043","version":1},"attestation_state":"computed","paper":{"title":"Towards Load-Aware Prefill Deflection for Disaggregated LLM Serving","license":"http://creativecommons.org/licenses/by-sa/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.DC","authors_text":"Anjaly Parayil, Renee St. Amant, Shrikara Arun, Srikant Bharadwaj, Victor R\\\"uhle","submitted_at":"2026-07-02T11:10:05Z","abstract_excerpt":"Disaggregated LLM serving runs prefill and decode on separate GPU pools to keep the two phases from interfering. In practice, this creates a new asymmetry: under bursty, heavy-tailed workloads prefill nodes saturate while decode nodes have compute underutilized, and on a production-style A100 cluster with 2 prefill and 2 decode nodes (2P2D), we find that prefill execution accounts for only 2-23% of P95 Time-to-First-Token (TTFT). Queuing and inter-node GPU-GPU KV-cache transfer account for the rest. We present a proactive prefill-deflecting scheduler that lets decode nodes serve prefill phase "},"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":"2607.02043","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by-sa/4.0/","primary_cat":"cs.DC","submitted_at":"2026-07-02T11:10:05Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"c3dd7d09c4d83fcb9f7065363c85a6688bd4033df8156aadb7f53df76ae9410b","abstract_canon_sha256":"3a7542b6c34cb2aa2a0a852d5d7cfabe7cb3208c2e9669e09a8cdd6e5ca66b50"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-03T01:17:37.906634Z","signature_b64":"gsl+PpULKoM6mZzB6dPBu63NGeqA1COHJdOxgid1uaXsWuUfpzYW//BQq0Jo3m3k6hTXECQUEfMOOldQUY8jDQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"d6f9d47d16a58d56da14d69802d039a3f6e65609fa4ff3e59ffdfb2a99d01155","last_reissued_at":"2026-07-03T01:17:37.906106Z","signature_status":"signed_v1","first_computed_at":"2026-07-03T01:17:37.906106Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Towards Load-Aware Prefill Deflection for Disaggregated LLM Serving","license":"http://creativecommons.org/licenses/by-sa/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.DC","authors_text":"Anjaly Parayil, Renee St. Amant, Shrikara Arun, Srikant Bharadwaj, Victor R\\\"uhle","submitted_at":"2026-07-02T11:10:05Z","abstract_excerpt":"Disaggregated LLM serving runs prefill and decode on separate GPU pools to keep the two phases from interfering. In practice, this creates a new asymmetry: under bursty, heavy-tailed workloads prefill nodes saturate while decode nodes have compute underutilized, and on a production-style A100 cluster with 2 prefill and 2 decode nodes (2P2D), we find that prefill execution accounts for only 2-23% of P95 Time-to-First-Token (TTFT). Queuing and inter-node GPU-GPU KV-cache transfer account for the rest. We present a proactive prefill-deflecting scheduler that lets decode nodes serve prefill phase "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2607.02043","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/2607.02043/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":"2607.02043","created_at":"2026-07-03T01:17:37.906176+00:00"},{"alias_kind":"arxiv_version","alias_value":"2607.02043v1","created_at":"2026-07-03T01:17:37.906176+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2607.02043","created_at":"2026-07-03T01:17:37.906176+00:00"},{"alias_kind":"pith_short_12","alias_value":"2345I7IWUWGV","created_at":"2026-07-03T01:17:37.906176+00:00"},{"alias_kind":"pith_short_16","alias_value":"2345I7IWUWGVNWQU","created_at":"2026-07-03T01:17:37.906176+00:00"},{"alias_kind":"pith_short_8","alias_value":"2345I7IW","created_at":"2026-07-03T01:17:37.906176+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":0,"internal_anchor_count":0,"sample":[]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/2345I7IWUWGVNWQU22MAFUBZUP","json":"https://pith.science/pith/2345I7IWUWGVNWQU22MAFUBZUP.json","graph_json":"https://pith.science/api/pith-number/2345I7IWUWGVNWQU22MAFUBZUP/graph.json","events_json":"https://pith.science/api/pith-number/2345I7IWUWGVNWQU22MAFUBZUP/events.json","paper":"https://pith.science/paper/2345I7IW"},"agent_actions":{"view_html":"https://pith.science/pith/2345I7IWUWGVNWQU22MAFUBZUP","download_json":"https://pith.science/pith/2345I7IWUWGVNWQU22MAFUBZUP.json","view_paper":"https://pith.science/paper/2345I7IW","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2607.02043&json=true","fetch_graph":"https://pith.science/api/pith-number/2345I7IWUWGVNWQU22MAFUBZUP/graph.json","fetch_events":"https://pith.science/api/pith-number/2345I7IWUWGVNWQU22MAFUBZUP/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/2345I7IWUWGVNWQU22MAFUBZUP/action/timestamp_anchor","attest_storage":"https://pith.science/pith/2345I7IWUWGVNWQU22MAFUBZUP/action/storage_attestation","attest_author":"https://pith.science/pith/2345I7IWUWGVNWQU22MAFUBZUP/action/author_attestation","sign_citation":"https://pith.science/pith/2345I7IWUWGVNWQU22MAFUBZUP/action/citation_signature","submit_replication":"https://pith.science/pith/2345I7IWUWGVNWQU22MAFUBZUP/action/replication_record"}},"created_at":"2026-07-03T01:17:37.906176+00:00","updated_at":"2026-07-03T01:17:37.906176+00:00"}