pith:J2QAIDUV
SAGA: Workflow-Atomic Scheduling for AI Agent Inference on GPU Clusters
Scheduling AI agent workflows as single units reduces task completion time by 1.64x on GPU clusters
arxiv:2605.00528 v2 · 2026-05-01 · cs.DC · cs.AI · cs.LG · cs.OS
Add to your LaTeX paper
\usepackage{pith}
\pithnumber{J2QAIDUV7JSXVG22VIX4DVT76T}
Prints a linked badge after your title and injects PDF metadata. Compiles on arXiv. Learn more · Embed verified badge
Record completeness
Claims
On a 64-GPU cluster serving SWE-bench coding agents and WebArena browser tasks, SAGA reduces task completion time by 1.64x (geometric mean, p < 0.001) over vLLM v0.15.1 with prefix caching and affinity routing, while improving GPU memory utilization by 1.22x and achieving 99.2% SLO attainment under multi-tenant interference.
That real-world agent workflows possess enough predictable structure for Agent Execution Graphs to accurately forecast KV cache reuse across tool-call boundaries, and that the dominant use case is latency-sensitive interactive work where a 30% throughput reduction is acceptable.
SAGA reduces AI agent task completion time by 1.64x on 64-GPU clusters by scheduling at the full workflow level with execution graphs, affinity batching, and completion-time fairness.
Receipt and verification
| First computed | 2026-06-23T01:12:07.813331Z |
|---|---|
| Builder | pith-number-builder-2026-05-17-v1 |
| Signature | Pith Ed25519
(pith-v1-2026-05) · public key |
| Schema | pith-number/v1.0 |
Canonical hash
4ea0040e95fa657a9b5aaa2fc1d67ff4c6b0255c65fad3a8c048cc30ad808bbf
Aliases
· · · · ·Agent API
Verify this Pith Number yourself
curl -sH 'Accept: application/ld+json' https://pith.science/pith/J2QAIDUV7JSXVG22VIX4DVT76T \
| jq -c '.canonical_record' \
| python3 -c "import sys,json,hashlib; b=json.dumps(json.loads(sys.stdin.read()), sort_keys=True, separators=(',',':'), ensure_ascii=False).encode(); print(hashlib.sha256(b).hexdigest())"
# expect: 4ea0040e95fa657a9b5aaa2fc1d67ff4c6b0255c65fad3a8c048cc30ad808bbf
Canonical record JSON
{
"metadata": {
"abstract_canon_sha256": "4ee36fde351ce84edcf098394946a2a402b13395be3b520a29cca5a99fa7ae6b",
"cross_cats_sorted": [
"cs.AI",
"cs.LG",
"cs.OS"
],
"license": "http://arxiv.org/licenses/nonexclusive-distrib/1.0/",
"primary_cat": "cs.DC",
"submitted_at": "2026-05-01T09:05:28Z",
"title_canon_sha256": "29c1169a251f1f9b1d2f4ce3ac03649d8a0ad6b41c66274100ccb236ea15aa1b"
},
"schema_version": "1.0",
"source": {
"id": "2605.00528",
"kind": "arxiv",
"version": 2
}
}