Link Search Menu Expand Document Documentation Menu
Documentation

Memory-optimized search

Introduced 3.1

Memory-optimized search allows the Faiss engine to run efficiently without loading the entire vector index into off-heap memory. Without this optimization, Faiss typically loads the full index into memory, which can become unsustainable if the index size exceeds available physical memory. With memory-optimized search, the engine memory-maps the index file and relies on the operating system’s file cache to serve search requests. This approach avoids unnecessary I/O and allows repeated reads to be served directly from the system cache.

Memory-optimized search affects only search operations. Indexing behavior remains unchanged.

Limitations

The following limitations apply to memory-optimized search in OpenSearch:

  • For indexes created before OpenSearch 2.19, the engine loads data into memory regardless of whether memory-optimized mode is enabled.
  • Memory-optimized search is supported only for the Faiss engine with the HNSW method.
  • Memory-optimized search does not support IVF or product quantization (PQ).
  • An index restart is required to enable or disable memory-optimized search.

If you use IVF or PQ, the engine loads data into memory regardless of whether memory-optimized mode is enabled.

Configuration

To enable memory-optimized search, set index.knn.memory_optimized_search to true when creating an index:

PUT /test_index { "settings": { "index.knn": true, "index.knn.memory_optimized_search": true }, "mappings": { "properties": { "vector_field": { "type": "knn_vector", "dimension": 128, "method": { "name": "hnsw", "engine": "faiss" } } } } }

To enable memory-optimized search on an existing index, you must close the index, update the setting, and then reopen the index:

POST /test_index/_close 

PUT /test_index/_settings { "index.knn.memory_optimized_search": true } 

POST /test_index/_open

When you configure a field with on_disk mode and 1x compression, memory-optimized search is automatically enabled for that field, even if memory optimization isn’t enabled at the index level. For more information, see Memory-optimized vectors.

Memory-optimized search differs from disk-based search because it doesn’t use compression or quantization. It only changes how vector data is loaded and accessed during search.

Performance optimization

When memory-optimized search is enabled, the warm-up API loads only the essential information needed for search operations, such as opening streams to the underlying Faiss index file. This minimal warm-up results in:

  • Faster initial searches.
  • Reduced memory overhead.
  • More efficient resource utilization.

For fields where memory-optimized search is disabled, the warm-up process loads vectors into off-heap memory.

Next steps

350 characters left

Have a question? .

Want to contribute? or .