Vulcan: Instance-specialized, Verifiable Systems Heuristics Through LLM-driven Search

Vulcan is a framework that leverages large language models (LLMs) to synthesize instance-specialized systems heuristics, addressing the challenges of hardware heterogeneity and workload diversity. It utilizes a restricted language called Anvil to generate safe, stateless decision functions, ensuring correct system integration while achieving significant performance improvements, such as up to 4.9x savings in spot-VM scheduling and a 10% increase in application performance for tiered-memory systems. This approach is crucial for practitioners as it automates heuristic generation, enhancing scalability and reliability in systems resource management.