Abstract
Reactive autoscaling systems may cause performance degradation and thermal issues when deploying new code, as such systems often scale resources based on lagging indicators. A system can address this by integrating a large language model (LLM) into a continuous integration/continuous deployment pipeline. Before deployment, the system analyzes the semantic and structural impact of codebase modifications by processing code diffs and abstract syntax tree deltas, which can be augmented with historical telemetry data. The LLM then generates a resource delta prediction vector to quantify anticipated physical demands, such as central processing unit cycles and thermal output. This prediction may enable a physical infrastructure control plane to proactively actuate hardware resources, for example, powering on servers or adjusting cooling, in parallel with the software build. This approach may help mitigate cold-starts and improve stability upon deployment.
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.
Recommended Citation
Yakar, Tamar and Labzovsky, Ilia, "Predictive Physical Hardware Actuation via LLM-Based Semantic Code Analysis", Technical Disclosure Commons, ()
https://www.tdcommons.org/dpubs_series/10027