Milvus is a

distributed Distribution may refer to: Mathematics *Distribution (mathematics), generalized functions used to formulate solutions of partial differential equations *Probability distribution, the probability of a particular value or value range of a varia ...

vector database A vector database, vector store or vector search engine is a database that uses the vector space model to store vectors (fixed-length lists of numbers) along with other data items. Vector databases typically implement one or more Nearest neighbor ...

developed by Zilliz. It is available as both

open-source Open source is source code that is made freely available for possible modification and redistribution. Products include permission to use and view the source code, design documents, or content of the product. The open source model is a decentrali ...

software and a

cloud service Cloud computing is "a paradigm for enabling network access to a scalable and elastic pool of shareable physical or virtual resources with self-service provisioning and administration on-demand," according to ISO. Essential characteristics ...

. Milvus is an open-source project under LF AI & Data Foundation distributed under the Apache License 2.0.

History

Milvus has been developed by Zilliz since 2017. Milvus joined

Linux Foundation The Linux Foundation (LF) is a non-profit organization established in 2000 to support Linux development and open-source software projects. Background The Linux Foundation started as Open Source Development Labs in 2000 to standardize and prom ...

as an incubation project in and became a graduate in . The details about its architecture and possible applications were presented on

ACM SIGMOD Conference SIGMOD is the Association for Computing Machinery's Special Interest Group on Management of Data, which specializes in large-scale data management problems and databases. The annual ACM SIGMOD Conference, which began in 1975, is considered one of ...

in 2021 Milvus 2.0, a major redesign of the whole product with a new architecture, was released in .

Features

Similarity search

Major similarity search related features that are available in the active 2.4.x Milvus branch: * In-memory, on-disk and GPU indices, * Single query, batch query and range query search, * Support of sparse vectors, binary vectors,

JSON JSON (JavaScript Object Notation, pronounced or ) is an open standard file format and electronic data interchange, data interchange format that uses Human-readable medium and data, human-readable text to store and transmit data objects consi ...

and arrays, * FP32,

FP16 In computing, half precision (sometimes called FP16 or float16) is a binary floating-point computer number format that occupies 16 bits (two bytes in modern computers) in computer memory. It is intended for storage of floating-point values in a ...

and

BF16 The bfloat16 (brain floating point) floating-point format is a computer number format occupying 16 bits in computer memory; it represents a wide dynamic range of numeric values by using a floating radix point. This format is a shortened (16-bi ...

data types, *

Euclidean distance In mathematics, the Euclidean distance between two points in Euclidean space is the length of the line segment between them. It can be calculated from the Cartesian coordinates of the points using the Pythagorean theorem, and therefore is o ...

, inner product distance and cosine distance support for floating-point data, *

Hamming distance In information theory, the Hamming distance between two String (computer science), strings or vectors of equal length is the number of positions at which the corresponding symbols are different. In other words, it measures the minimum number ...

and

jaccard distance The Jaccard index is a statistic used for gauging the similarity and diversity of sample sets. It is defined in general taking the ratio of two sizes (areas or volumes), the intersection size divided by the union size, also called intersection ...

for

binary data Binary data is data whose unit can take on only two possible states. These are often labelled as 0 and 1 in accordance with the binary numeral system and Boolean algebra. Binary data occurs in many different technical and scientific fields, wh ...

, * Support of graph indices (including HNSW), Inverted-lists based indices and a

brute-force search In computer science, brute-force search or exhaustive search, also known as generate and test, is a very general problem-solving technique and algorithmic paradigm that consists of Iteration#Computing, systematically checking all possible candida ...

. * Support of

vector quantization Vector quantization (VQ) is a classical quantization technique from signal processing that allows the modeling of probability density functions by the distribution of prototype vectors. Developed in the early 1980s by Robert M. Gray, it was ori ...

for lossy input data compression, including

product quantization Product may refer to: Business * Product (business), an item that can be offered to a market to satisfy the desire or need of a customer. * Product (project management), a deliverable or set of deliverables that contribute to a business solution ...

(PQ) and scalar quantization (SQ), that trades stored data size for accuracy, * Re-ranking. Milvus similarity search engine relies on heavily-modified

forks In cutlery or kitchenware, a fork (from 'pitchfork') is a Eating utensil, utensil, now usually made of metal, whose long handle terminates in a head that branches into several narrow and often slightly curved tine (structural), tines with whic ...

of third-party open-source similarity search libraries, such as

Faiss FAISS (Facebook AI Similarity Search) is an open-source library for similarity search and clustering of vectors. It contains algorithms that search in sets of vectors of any size, up to ones that possibly do not fit in RAM. It also contains su ...

, DiskANN and hnswlib. Milvus includes optimizations for I/O data layout, specific to graph search indices.

Database

As a

database In computing, a database is an organized collection of data or a type of data store based on the use of a database management system (DBMS), the software that interacts with end users, applications, and the database itself to capture and a ...

, Milvus provides the following features: *

Column-oriented database Data orientation is the representation of tabular data in a linear memory model such as in-disk or in-memory. The two most common representations are column-oriented (columnar format) and row-oriented (row format). The choice of data orientat ...

* Four supported

data consistency Data inconsistency refers to whether the same data kept at different places do or do not match. Point-in-time consistency Point-in-time consistency is an important property of backup files and a critical objective of software that creates backups. ...

levels, including strong consistency and

eventual consistency Eventual consistency is a consistency model used in distributed computing to achieve high availability. Put simply: if no new updates are made to a given data item, ''eventually'' all accesses to that item will return the last updated value. Eve ...

. * Data sharding * Streaming data ingestion, which allows to process and ingest data in real-time as it arrives * A dynamic schema, which allows inserting the data without a predefined schema * Independent storage and compute layers *

Multi-tenancy Software multitenancy is a software architecture in which a single instance of software runs on a server and serves multiple tenants. Systems designed in such manner are "shared" (rather than "dedicated" or "isolated"). A tenant is a group of us ...

scenarios (database-oriented, collection-oriented, partition-oriented) * Memory-mapped data storage *

Role-based access control In computer systems security, role-based access control (RBAC) or role-based security is an approach to restricting system access to authorized users, and to implementing mandatory access control (MAC) or discretionary access control, discretion ...

* Multi-vector and hybrid search

Deployment options

Milvus can be deployed as an embedded database, standalone server, or distributed cluster. Zillis Cloud offers a fully managed version.

GPU support

Milvus provides GPU accelerated index building and search using

Nvidia Nvidia Corporation ( ) is an American multinational corporation and technology company headquartered in Santa Clara, California, and incorporated in Delaware. Founded in 1993 by Jensen Huang (president and CEO), Chris Malachowsky, and Curti ...

CUDA In computing, CUDA (Compute Unified Device Architecture) is a proprietary parallel computing platform and application programming interface (API) that allows software to use certain types of graphics processing units (GPUs) for accelerated gene ...

technology via Nvidia RAFT library, including a recent GPU-based graph indexing algorithm Nvidia CAGRA

Integration

Milvus provides official SDK clients for

Java Java is one of the Greater Sunda Islands in Indonesia. It is bordered by the Indian Ocean to the south and the Java Sea (a part of Pacific Ocean) to the north. With a population of 156.9 million people (including Madura) in mid 2024, proje ...

, NodeJS, Python and Go. An additional C# SDK client was contributed by

Microsoft Microsoft Corporation is an American multinational corporation and technology company, technology conglomerate headquartered in Redmond, Washington. Founded in 1975, the company became influential in the History of personal computers#The ear ...

. The database can integrate with

Prometheus In Greek mythology, Prometheus (; , , possibly meaning "forethought")Smith"Prometheus". is a Titans, Titan. He is best known for defying the Olympian gods by taking theft of fire, fire from them and giving it to humanity in the form of technol ...

and Grafana for monitoring and alerts, frameworks Haystack and LangChain,

IBM Watsonx Watsonx is IBM's commercial generative AI and scientific data platform based on cloud computing, cloud. It offers a studio, data store, and governance toolkit. It supports multiple large language models (LLMs) along with IBM's own IBM Granite ...

, and OpenAI models.

References

{{Reflist, 30em Software using the Apache license Vector databases