{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:7Z62GUQFPE4EE6LGC3POT2YUMR","short_pith_number":"pith:7Z62GUQF","schema_version":"1.0","canonical_sha256":"fe7da35205793842796616dee9eb14647a7b849d3aeeba3fd5ac2dbca31611c1","source":{"kind":"arxiv","id":"2605.16637","version":1},"attestation_state":"computed","paper":{"title":"HexAGenT: Efficient Agentic LLM Serving via Workflow- and Heterogeneity-Aware Scheduling","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"HexAGenT schedules agentic LLM workflows on heterogeneous GPU clusters to cut the SLO scale needed for timely end-to-end completion.","cross_cats":[],"primary_cat":"cs.DC","authors_text":"Binhang Yuan, Chen Wang, Jiawei Jiang, Ke Zhou, Wenshuang Li, Xu Xu, Youhe Jiang, You Peng","submitted_at":"2026-05-15T21:09:34Z","abstract_excerpt":"Agentic LLM applications increasingly execute user requests as multi-step workflows involving planning, tool use, branching, refinement, and synthesis. In such settings, users experience the end-to-end latency of an entire workflow, not the latency of any single LLM call. In this paper, we study how to schedule online agentic workflows across heterogeneous prefill-decode disaggregated LLM serving clusters to efficiently meet workflow-level latency objectives. The problem is challenging because workflow dependencies are revealed incrementally at runtime, calls have heterogeneous prompts, output"},"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":true,"formal_links_present":true},"canonical_record":{"source":{"id":"2605.16637","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.DC","submitted_at":"2026-05-15T21:09:34Z","cross_cats_sorted":[],"title_canon_sha256":"98083d893fbed3e51d3fd44140d85c745654be7d1d10774a8ce1d3b9a6baf02c","abstract_canon_sha256":"77d375aa47b489130bc7f7a1d563f69c88050fe4ec8cbdc80b55a67c2779489a"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-20T00:02:33.662765Z","signature_b64":"v7Egno4wNVAmU2cEJBRGg1jp4alSZHNp/ENew1uXyvKM4c28zkJPxsJnLldRN9Qe1JBLp8f4ecuqiA0YiWE9CQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"fe7da35205793842796616dee9eb14647a7b849d3aeeba3fd5ac2dbca31611c1","last_reissued_at":"2026-05-20T00:02:33.661896Z","signature_status":"signed_v1","first_computed_at":"2026-05-20T00:02:33.661896Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"HexAGenT: Efficient Agentic LLM Serving via Workflow- and Heterogeneity-Aware Scheduling","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"HexAGenT schedules agentic LLM workflows on heterogeneous GPU clusters to cut the SLO scale needed for timely end-to-end completion.","cross_cats":[],"primary_cat":"cs.DC","authors_text":"Binhang Yuan, Chen Wang, Jiawei Jiang, Ke Zhou, Wenshuang Li, Xu Xu, Youhe Jiang, You Peng","submitted_at":"2026-05-15T21:09:34Z","abstract_excerpt":"Agentic LLM applications increasingly execute user requests as multi-step workflows involving planning, tool use, branching, refinement, and synthesis. In such settings, users experience the end-to-end latency of an entire workflow, not the latency of any single LLM call. In this paper, we study how to schedule online agentic workflows across heterogeneous prefill-decode disaggregated LLM serving clusters to efficiently meet workflow-level latency objectives. The problem is challenging because workflow dependencies are revealed incrementally at runtime, calls have heterogeneous prompts, output"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"Across representative agentic workloads and heterogeneous A100/H100/H200 clusters, HexAGenT reduces the SLO scale required for timely workflow completion by an average of 20.1% at 95% attainment and 33.0% at 99% attainment, with maximum reductions of 45.0% and 80.5%, respectively.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"The representative agentic workloads and the specific heterogeneous A100/H100/H200 cluster configurations used for evaluation are sufficiently similar to real production deployments that the reported reductions in required SLO scale will generalize.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"HexAGenT reduces the SLO scale required for timely agentic LLM workflow completion by an average of 20.1% at 95% attainment and 33.0% at 99% attainment on heterogeneous A100/H100/H200 clusters.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"HexAGenT schedules agentic LLM workflows on heterogeneous GPU clusters to cut the SLO scale needed for timely end-to-end completion.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"3ce4cf639c6e1c57c4d7dca18f1d58e36de4c563be4f1a7cedeaf3db0065b039"},"source":{"id":"2605.16637","kind":"arxiv","version":1},"verdict":{"id":"2f5f7b21-eeb8-4912-a455-a1b8b2d2d74d","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-19T20:55:31.072816Z","strongest_claim":"Across representative agentic workloads and heterogeneous A100/H100/H200 clusters, HexAGenT reduces the SLO scale required for timely workflow completion by an average of 20.1% at 95% attainment and 33.0% at 99% attainment, with maximum reductions of 45.0% and 80.5%, respectively.","one_line_summary":"HexAGenT reduces the SLO scale required for timely agentic LLM workflow completion by an average of 20.1% at 95% attainment and 33.0% at 99% attainment on heterogeneous A100/H100/H200 clusters.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"The representative agentic workloads and the specific heterogeneous A100/H100/H200 cluster configurations used for evaluation are sufficiently similar to real production deployments that the reported reductions in required SLO scale will generalize.","pith_extraction_headline":"HexAGenT schedules agentic LLM workflows on heterogeneous GPU clusters to cut the SLO scale needed for timely end-to-end completion."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2605.16637/integrity.json","findings":[],"available":true,"detectors_run":[{"name":"doi_compliance","ran_at":"2026-05-19T21:02:02.885286Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"doi_title_agreement","ran_at":"2026-05-19T21:01:19.322449Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"claim_evidence","ran_at":"2026-05-19T19:21:56.715218Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"ai_meta_artifact","ran_at":"2026-05-19T18:33:26.577638Z","status":"skipped","version":"1.0.0","findings_count":0}],"snapshot_sha256":"9703c1bea1a6e4cb3ba75f12ffff5280d26647ef8e798fd32812a93e68b01886"},"references":{"count":57,"sample":[{"doi":"","year":2024,"title":"Gulavani, Alexey Tumanov, and Ramachandran Ramjee","work_id":"58c260ce-fb5f-42ba-9561-280d74ae76c8","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"10.1145/195058.195412","year":1994,"title":"Yossi Azar, Andrei Z. Broder, Anna R. Karlin, and Eli Upfal. Balanced allocations. In Proceedings of the Twenty-Sixth Annual ACM Symposium on Theory of Computing, pages 593–602, New York, NY, USA, 199","work_id":"aeb4e9a9-f2a9-4067-bf6d-bcd600cf1997","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2024,"title":"Graph of thoughts: Solving elaborate problems with large language models","work_id":"9a9e6a5f-db15-459f-8bd7-40118f0d8c19","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"10.1145/3669940.3707267","year":2025,"title":"Gonzalez, Matei Zaharia, and Ion Stoica","work_id":"9c5c0b40-d793-45bb-9da5-92d3d64d8555","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2023,"title":"Gonzalez, Ion Stoica, and Eric P","work_id":"0eb618ac-525f-481d-8b74-40c468fade71","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":57,"snapshot_sha256":"ecc21997458aad26295f38052f00c7b561744f09ee142a93752824d0d26cb104","internal_anchors":3},"formal_canon":{"evidence_count":2,"snapshot_sha256":"d31453f7122f4d4580a4802e52f02869b7cb824c4dad0b27ad0cc6753eb205fb"},"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":"2605.16637","created_at":"2026-05-20T00:02:33.662055+00:00"},{"alias_kind":"arxiv_version","alias_value":"2605.16637v1","created_at":"2026-05-20T00:02:33.662055+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.16637","created_at":"2026-05-20T00:02:33.662055+00:00"},{"alias_kind":"pith_short_12","alias_value":"7Z62GUQFPE4E","created_at":"2026-05-20T00:02:33.662055+00:00"},{"alias_kind":"pith_short_16","alias_value":"7Z62GUQFPE4EE6LG","created_at":"2026-05-20T00:02:33.662055+00:00"},{"alias_kind":"pith_short_8","alias_value":"7Z62GUQF","created_at":"2026-05-20T00:02:33.662055+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":0,"internal_anchor_count":0,"sample":[]},"formal_canon":{"evidence_count":2,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/7Z62GUQFPE4EE6LGC3POT2YUMR","json":"https://pith.science/pith/7Z62GUQFPE4EE6LGC3POT2YUMR.json","graph_json":"https://pith.science/api/pith-number/7Z62GUQFPE4EE6LGC3POT2YUMR/graph.json","events_json":"https://pith.science/api/pith-number/7Z62GUQFPE4EE6LGC3POT2YUMR/events.json","paper":"https://pith.science/paper/7Z62GUQF"},"agent_actions":{"view_html":"https://pith.science/pith/7Z62GUQFPE4EE6LGC3POT2YUMR","download_json":"https://pith.science/pith/7Z62GUQFPE4EE6LGC3POT2YUMR.json","view_paper":"https://pith.science/paper/7Z62GUQF","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2605.16637&json=true","fetch_graph":"https://pith.science/api/pith-number/7Z62GUQFPE4EE6LGC3POT2YUMR/graph.json","fetch_events":"https://pith.science/api/pith-number/7Z62GUQFPE4EE6LGC3POT2YUMR/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/7Z62GUQFPE4EE6LGC3POT2YUMR/action/timestamp_anchor","attest_storage":"https://pith.science/pith/7Z62GUQFPE4EE6LGC3POT2YUMR/action/storage_attestation","attest_author":"https://pith.science/pith/7Z62GUQFPE4EE6LGC3POT2YUMR/action/author_attestation","sign_citation":"https://pith.science/pith/7Z62GUQFPE4EE6LGC3POT2YUMR/action/citation_signature","submit_replication":"https://pith.science/pith/7Z62GUQFPE4EE6LGC3POT2YUMR/action/replication_record"}},"created_at":"2026-05-20T00:02:33.662055+00:00","updated_at":"2026-05-20T00:02:33.662055+00:00"}