Milvus

Milvus is a cloud-native, open-source vector database that allows you to store, index, and manage massive embedding vectors generated by deep neural networks and other machine learning (ML) models. It is built to power embedding similarity search and AI applications, making unstructured data search more accessible, and providing a consistent user experience regardless of the deployment environment. Milvus 2.0 is a cloud-native vector database with storage and computation separated by design.

You can deploy Milvus in your Nebius AI Managed Service for Kubernetes clusters using this Marketplace product.

Milvus supports GPU acceleration that uses NVIDIA GPUs to sustain AI workflows. In this product, GPU acceleration is not enabled by default, and you can enable it after installing Milvus. For details, see deployment instructions below.

Before installing this product:

1. Create a Kubernetes cluster and a node group in it. If you want to use GPU acceleration, the node group must contain VMs with GPUs.

2. Install kubectl and configure it to work with the created cluster.

3. To install the product:

   • On the cluster page in the management console, go to the Marketplace tab, select the product, and click Install.

4. Configure the application:

   • Namespace: Select a namespace or create a new one.

   • Application name: Enter an application name.

   • GPU number: Enter the number of GPUs for GPU acceleration. If the number is positive, the node group in the cluster must contain at least this number of VMs with GPUs. If the value is 0 or empty, Milvus will run on vCPUs only.

   • Enable Cluster mode: Select this option to install Milvus as a Kubernetes operator, which allows you to manage a full Milvus service stack. By default, Milvus is installed standalone as a Docker container. For details, see the Milvus documentation about Milvus Cluster and Milvus Standalone.

   • Enable Pulsar: If the cluster mode is enabled, select this option to install Apache Pulsar^™, the messaging solution that supports Milvus’s storage and handling of message streams.

   • Number of etcd replicas: Enter the number of etcd replicas that should be created for Milvus.

   • Minio mode: Select a deployment type for MinIO, the storage engine used by Milvus:

     • standalone: The MinIO Server runs on a single node in a single server pool. Recommended only for testing environments with Milvus in standalone mode.
     • distributed: The MinIO Server runs in one or multiple multi-node server pool. Recommended for production environments or for using Milvus in cluster mode.

     For more details, see the MinIO documentation.

5. Click Install.

6. Wait for the application to change its status to Deployed.

7. To check that Milvus is working, check that its pods are running:

   kubectl get pods -n <namespace> | grep "milvus-operator"
   

• Image and video similarity search.
• Natural language processing (NLP).
• Building personalized recommendation systems.
• Storing and retrieving genomic vectors for genetic data analysis, comparing similar medical images for diagnosis or research.
• Anomaly and fraud detection.
• Geospatial data indexing, rapid retrieval of similar locations, routes in mapping and navigation applications.
• Storing and retrieving embeddings of machine learning models.
• Analysis of financial time series for risk management, forecasting, and algorithmic trading.
• Augmented reality (AR) and virtual reality (VR), similarity search for 3D models, textures, or other spatial features.

The Nebius AI support team and architects can consult you about whether and how the product can be deployed for production use.
Nebius AI does not provide full technical support for the product. If you have any issues, please refer to the developer’s information resources.