Milvus: A Purpose-Built Vector Data Management System

Recently, there has been a pressing need to manage high-dimensional vector data in data science and AI applications. This trend is fueled by the proliferation of unstructured data and machine learning (ML), where ML models usually transform unstructured data into feature vectors for data analytics, e.g., product recommendation. Existing systems and algorithms for managing vector data have two limitations: (1) They incur serious performance issue when handling large-scale and dynamic vector data; and (2) They provide limited functionalities that cannot meet the requirements of versatile applications.This paper presents Milvus, a purpose-built data management system to efficiently manage large-scale vector data. Milvus supports easy-to-use application interfaces (including SDKs and RESTful APIs); optimizes for the heterogeneous computing platform with modern CPUs and GPUs; enables advanced query processing beyond simple vector similarity search; handles dynamic data for fast updates while ensuring efficient query processing; and distributes data across multiple nodes to achieve scalability and availability. We first describe the design and implementation of Milvus. Then we demonstrate the real-world use cases supported by Milvus. In particular, we build a series of 10 applications on top of Milvus. Finally, we experimentally evaluate Milvus with a wide range of systems including ElasticSearch, Vearch, Alibaba PASE (PostgreSQL), Alibaba AnalyticDB-V, and Microsoft SPTAG. Experiments show that Milvus is up to two orders of magnitude faster than the competitors while providing more functionalities. Now Milvus is deployed by hundreds of organizations worldwide and it is also recognized as an incubation-stage project of the LF AI & Data Foundation. Milvus is open-sourced at https://github.com/milvus-io/milvus.