Closed-Loop Resource Scaling With Verifiable Safety Bounds via LLM-Generated Constraint Programs
Journal of Innovative Research and Technology · 25.06.2026 · 3 (1) · 187-199
DOI: 10.66857/jirt.v3i1.45 ↗Kurz gefasst
Fachartikel in der Zeitschrift Journal of Innovative Research and Technology, online erschienen am 25.06.2026. Verfasst von Ethan Parker und 1 weiteren Autor:innen. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).
Abstract (Originalsprache)
Modern cloud-native infrastructure demands resource scaling systems that respond dynamically to workload fluctuations while providing formal guarantees that safety constraints — including CPU utilization ceilings, memory limits, and service level objective (SLO) thresholds — are never violated. Existing autoscaling approaches rely on threshold-based rules or reinforcement learning (RL) policies that lack interpretable, verifiable safety properties. This paper proposes a novel closed-loop resource scaling framework in which a large language model (LLM) synthesizes constraint programs encoding system-specific safety bounds, and a feedback controller uses these programs to verify and govern scaling decisions at runtime. The generated constraint programs are expressed in a standard constraint satisfaction problem (CSP) representation and are formally verified before deployment. Experimental evaluations on representative microservice workloads in a Kubernetes environment demonstrate that the proposed framework reduces SLO violations by 41.3% and lowers peak resource overprovisioning by 28.7% compared to production-grade baselines, while maintaining mathematically verifiable safety bounds across all evaluated scenarios
Zitieren
APA
Parker, E., & Thompson, G. (2026). Closed-Loop Resource Scaling With Verifiable Safety Bounds via LLM-Generated Constraint Programs. Journal of Innovative Research and Technology, 3(1), 187-199. https://doi.org/10.66857/jirt.v3i1.45
Vancouver
Parker E, Thompson G. Closed-Loop Resource Scaling With Verifiable Safety Bounds via LLM-Generated Constraint Programs. Journal of Innovative Research and Technology. 2026;3(1):187-199. doi:10.66857/jirt.v3i1.45
BibTeX
@article{Parker2026,
author = {Parker, Ethan and Thompson, Grace},
title = {Closed-Loop Resource Scaling With Verifiable Safety Bounds via LLM-Generated Constraint Programs},
journal = {Journal of Innovative Research and Technology},
year = {2026},
volume = {3},
number = {1},
pages = {187--199},
doi = {10.66857/jirt.v3i1.45}
}
EUVIMED