Qdrant has announced the release of Qdrant 1.16, an important update to its open-source vector database. This new version introduces tiered multitenancy, a feature designed to improve tenant isolation, boost performance, and enable more efficient scaling of search workloads. Additionally, Qdrant 1.16 includes the innovative ACORN search algorithm, which enhances the accuracy of filtered vector searches, particularly when multiple filters with weak selectivity are used.

Enhanced Multitenancy with Tiered Architecture

The tiered multitenancy feature allows users to better manage multiple tenants within a single collection by supporting user-defined sharding, fallback shards, and tenant promotion. This approach enables small tenants to share resources efficiently while providing the ability to promote larger tenants to dedicated shards as they grow, thereby improving overall performance and isolation.

Implementing tiered multitenancy involves creating named shards for large tenants and routing requests through fallback shards when necessary. When a tenant’s usage increases, they can be promoted from shared fallback shards to dedicated shards, ensuring consistent performance and resource allocation. This flexible model aligns well with SaaS applications sharing a common database infrastructure.

Introduction of ACORN Search Algorithm

Qdrant’s new ACORN (Approximate Nearest Neighbor Constraint-Optimized Retrieval Network) algorithm significantly improves filtered vector search quality. It enhances the traversal of the HNSW (Hierarchical Navigable Small World) graph by examining neighbors of neighbors when direct neighbors are filtered out, leading to higher recall and more accurate search results.

While ACORN provides better accuracy, it does so at a performance cost—typically slowing searches by 2 to 10 times in common scenarios. Users can decide when to enable ACORN based on their accuracy needs versus performance constraints. Qdrant offers a decision matrix to guide users in making this choice effectively.

In addition to these core features, Qdrant 1.16 introduces a revamped user interface designed to enhance usability. The new UI includes a welcome page with quick access to tutorials and documentation, along with redesigned views for Points, Visualization, and Graphs within the Collections Manager, making data management more intuitive and streamlined.

Inspired by

• https://www.infoworld.com/article/4099002/qdrant-vector-database-adds-tiered-multitenancy.html

Sources

• qdrant.tech
• qdrant.tech
• arxiv.org
• qdrant.tech