Vulcan: Instance-specialized, Verifiable Systems Heuristics Through LLM-driven Search
Researchers have developed Vulcan, a novel framework that leverages Large Language Models (LLMs) to synthesize specialized system heuristics. This approach aims to address the challenges of manual heuristic design in increasingly diverse hardware and workload environments. Vulcan isolates core decision logic, restricting LLM-generated code to simple stateless functions and ensuring safety through a restricted language called Anvil. Evaluations show significant performance improvements, including up to 4.9x higher savings for spot-VM scheduling and up to 2x lower miss ratios for cache eviction. AI
IMPACT This research could enable more efficient and adaptable system resource management by automating the creation of specialized heuristics.