VDTuner
VDTuner is an automated performance tuning framework for Vector Data Management Systems, specifically demonstrated with Milvus and benchmarked using vector-db-benchmark. Presented at ICDE 2024, this tool optimizes search speed and recall rates by automatically configuring system parameters using Bayesian optimization via the BoTorch library. The system requires a Linux environment with Python 3.11, a deployed Milvus instance, and specific benchmark datasets like GloVe. VDTuner works by iteratively testing configuration combinations, adjusting engine parameters defined in JSON and YAML files, and evaluating performance against user-defined workload constraints and timeouts. It supports tuning various index types and parallel execution settings to adapt to specific hardware specifications. Designed for researchers and engineers, it eliminates manual trial-and-error in database configuration, providing a systematic approach to maximize vector database efficiency. The workflow involves preparing datasets, modifyi