迈向面向智能体的高效HPC系统：挑战与机遇
Towards Efficient HPC Systems for Agents: Challenges and Opportunities
- Harvard University（哈佛大学）
- Harvard FAS Research Computing（哈佛文理学院研究计算中心）
机构由 AI 辅助整理，请以论文原文为准。
中文总结 AI 辅助
本文针对编码智能体大量占用HPC资源的问题，提出将智能体视为一等公民的协同设计理念，并概述计算、存储、记忆和安全方面的挑战与机遇。
中文摘要 AI 辅助
编码智能体已成为高性能计算（HPC）系统的真实用户，然而当今的HPC抽象、接口和策略仍是为人工驱动的工作流设计的。在我们的测量中，运行编码智能体的用户仅占观测人口的19.5%，却贡献了55.8%的作业提交、29.1%的CPU核时和42.7%的GPU时。智能体并非仅仅是更快的用户。它们以人类速率20.8倍的速度发出命令，将工作分解为细粒度的探索-修改-执行循环，并通过持续到夜晚和周末的试错活动追求开放式目标。这些行为给为人类时间尺度构建的抽象带来了压力，表现为调度器上的控制平面压力、带宽配置文件系统上的元数据密集型I/O、对先前会话已学内容的重复发现，以及新的提示注入和策略执行面。既不能禁止智能体，也不能将其视为普通用户，这两种方式都不可持续。我们转而主张协同设计，即设施应将智能体视为一等公民，使其成为设施感知的租户。我们概述了由此在计算、存储、智能体记忆和安全方面产生的挑战与机遇。
英文摘要
Coding agents have become real users of high-performance computing (HPC) systems, yet today's HPC abstractions, interfaces, and policies remain designed for human-driven workflows. In our measurement, users running coding agents are only 19.5% of the observed population, but account for 55.8% of job submissions, 29.1% of CPU core-hours, and 42.7% of GPU-hours. Agents are not simply faster humans. They issue commands at 20.8x the human rate, decompose work into fine-grained explore-modify-execute loops, and pursue open-ended goals through trial-and-error campaigns that continue through nights and weekends. These behaviors strain abstractions built for human timescales, surfacing as control-plane pressure on the scheduler, metadata-intensive I/O on bandwidth-provisioned filesystems, repeated rediscovery of what earlier sessions already learned, and new prompt-injection and policy-enforcement surfaces. Neither banning agents nor treating them as ordinary users is sustainable. We instead argue for co-design, that facilities should treat agents as first-class principals where agents become facility-aware tenants. We outline the resulting challenges and opportunities in compute, storage, agent memory, and safety.
