Rollout 基建税差 110×：Daytona（产业 SoCC 2026），1M×150-step worker-hour 差 1.8×
Rollout 基建税差 110×：Daytona（产业 SoCC 2026），1M×150-step worker-hour 差 1.8×
View PDF HTML (experimental) Abstract: Coding-agent reinforcement learning treats execution infrastructure as a background implementation detail, despite relying on large numbers of interactive software rollouts. This is a missed opportunity: measuring infrastructure overhead...
Daytona（产业 SoCC 2026）发布 (arXiv:2607.01415)
“View PDF HTML (experimental) Abstract: Coding-agent reinforcement learning treats execution infrastructure as a background implementation de”
Abstract:Coding-agent reinforcement learning treats execution infrastructure as a background implementation detail, despite relying on large numbers of interactive software rollouts. This is a missed opportunity: measuring infrastructure overhead can reveal practical efficiency gains for RL post-training, where small per-rollout savings compound at scale. We present a comparative study of four execution substrates: single containers, hosted sandboxes, Kubernetes-orchestrated containers, and cloud virtual machines. We find up to $110\times$ variation in cold-start latency and a $1.8\times$ spread in projected worker-hours for one million 150-step trajectories. Our results suggest that future coding-agent RL systems should optimize execution substrates as part of the training system itself, not merely as deployment plumbing.
| Comments: | Preprint. 6 pages, 6 figures, 2 tables. Submitted to ACM SoCC 2026 |
| Subjects: | Machine Learning (cs.LG); Distributed, Parallel, and Cluster Computing (cs.DC) |
| ACM classes: | C.4; D.4.8; I.2.6 |
| Cite as: | arXiv:2607.01415 [cs.LG] |
| (or arXiv:2607.01415v1 [cs.LG] for this version) | |
| https://doi.org/10.48550/arXiv.2607.01415 arXiv-issued DOI via DataCite |
Submission history
From: Daniel Graviet [view email]
[v1]
Wed, 1 Jul 2026 19:20:46 UTC (214 KB)
Source: Agent Control Flash News · arxiv.org
Related Products & Analysis
Cursor Agent 架构全景拆解：从影子工作区到多文件推演引擎
全面剖析 Cursor 在 Agent 模式下的底层技术实现，解析影子分支隔离、LSP 上下文剪枝以及预测性补全的核心架构细节与工程实践经验。
Vercel v0 推出全新多 Agent 协作画布与实时全栈部署预览
动态速评：v0 正在从单一组件生成器演进为全栈应用搭建协作台，尤其是与 Neon/Postgres 的直连极大降低了全栈原型验证成本。