Optimizing vector storage

Vector search operations can be resource intensive, especially when dealing with large-scale vector datasets. OpenSearch provides several optimization techniques for reducing memory usage.

Disk-friendly quantization using the opensearch-jvector plugin

The jvector engine, provided by the opensearch-jvector plugin, implements DiskANN-style indexing: it stores vectors on disk rather than in memory and builds indexes directly from quantized vectors. This reduces memory use without requiring separate quantization configuration.

Compared with the built-in engines, the jvector engine provides the following storage advantages:

• It builds indexes from quantized vectors, reducing the memory required during indexing.
• It refines quantization codebooks incrementally during merges, with no full rebuilds.