Vector Search Performance Enhancements on Limited-Memory Edge Devices

Vector databases represent a significant advancement in data indexing and search technologies, enhancing the management of complex, unstructured data like multimedia. By encoding data into high-dimensional embeddings, they facilitate semantic-based retrieval, improving search relevance and user experience. Due to high computational demands, vector databases usually operate in cloud environments, making them impractical for resource-constrained edge computing. We present a vector storage and search system tailored for edge devices with significant computational and memory constraints. By reviewing existing vector search technologies and developing a proof-of-concept library for isolated edge environments, we demonstrate the potential for efficient vector search systems on edge devices. Our system achieves an average query time under 0.1 seconds and a recall rate exceeding 90%, even when handling million-scale, high-dimensional datasets on limited hardware.