Edge controlled system improves scientific computing with safer cloud AI help
ECAS: An Edge-Controlled Agentic System for Validation-Gated Scientific Application Execution
Distributed, Parallel, and Cluster ComputingArtificial Intelligence
Summary
Running big scientific programs on powerful computers can be tricky and often breaks if things don't match the computer’s specific setup. The authors created a system called ECAS that keeps the computer’s access codes safe on local devices, while a cloud AI suggests plans and fixes problems. The system checks small test runs before doing big jobs, so mistakes get fixed early without risking the full run. This approach lets AI help more safely and effectively, making science computing smoother with less human oversight.
What this means in practice
- •For scientific computing teams: Run complex science workflows on HPC systems with automated fault detection and repair controlled locally for safer execution.
- •For cloud security engineers: Use edge agents to protect sensitive credentials while enabling cloud-hosted AI to plan and fix computing workflows remotely.
Authors
Baixi Sun, Mingze Xia, Huihuo Zheng
Abstract
Scientific applications increasingly rely on high-performance computing (HPC), yet translating a scientist's high-level goal into a correct target-scale execution remains brittle and labor-intensive. Large language model (LLM) agents promise to automate this, but two obstacles remain: granting a cloud-hosted model direct HPC access exposes credentials and execution authority, while withholding it demands continuous human supervision; and one-shot generation cannot adapt when generated artifacts fail in a site-specific HPC environment. We present \textsc{ECAS}, an \textbf{E}dge-\textbf{C}ontrolled \textbf{A}gentic \textbf{S}ystem for closed-loop execution of scientific computing campaigns with limited human intervention. \textsc{ECAS} separates \emph{reasoning}, \emph{control}, and \emph{execution}: a cloud-hosted LLM proposes plans, artifacts, and repairs; a user-controlled edge agent retains credentials, workflow state, and execution authority while enforcing policy and resource constraints; and the HPC system computes. Its core mechanism is \emph{validation-gated execution}: generated artifacts pass static checks and small-scale validation, failures trigger repairs from sanitized execution feedback, and target-scale execution is permitted only after validation and policy checks pass. \textsc{ECAS} also draws on an edge-resident library of expert-distilled, site-specific skills that is never disclosed to the cloud. In preliminary experiments with three scientific applications on two production ALCF systems under six injected fault types, closed-loop repair improves application success from 0/6 to 6/6 over one-shot generation, validation gating prevents all three observed target-scale failures, and skill conditioning improves success from 4/6 to 6/6. These results show the feasibility of delegating adaptive reasoning to the cloud while retaining execution control at the edge.