Azure Cosmos DB for NoSQL introduces sharded DiskANN, a powerful feature enhancing your vector search experience. Sharding partitions a DiskANN vector index into smaller, more manageable segments. By simply defining the VectorIndexShardKey parameter in your collection’s indexing policy, you can effortlessly shard vector indexes based on your data’s attributes, significantly improving search efficiency and reducing costs.
Lower latency. Smaller, sharded indexes accelerate vector searches, ensuring faster response times for your applications.
Reduced costs. Smaller indexes mean less computational overhead, directly translating to lower Request Unit (RU) charges.
Enhanced accuracy and recall. Searches confined to focused subsets of your data improve accuracy and recall, delivering more relevant results by limiting the search scope to specific shard keys.
Effortless scalability. Sharding distributes your data efficiently, making it straightforward to scale your vector search capability as your dataset grows, ensuring consistent performance without complexity.
Data isolation and security. By isolating vector data based on shard keys, sharded DiskANN provides enhanced data management and security, restricting searches to relevant data subsets.