Optimizer Plans for Vector Indexes

Optimizer plans for HNSW and IVF indexes are described in the following sections.

• Optimizer Plans for HNSW Vector Indexes
  A Hierarchical Navigable Small World Graph (HNSW) is a form of In-Memory Neighbor Graph vector index. It is a very efficient index for vector approximate similarity search.
• Optimizer Plans for Local HNSW Vector Indexes
  Oracle AI Database uses partition pruning to narrow the search and search only the local HNSW graphs for qualifying partitions or sub partitions. Partition pruning narrows the set of local HNSW graphs that need to be searched, and the optimizer plan determines the most efficient execution plan for accessing narrowed graph set during similarity search. When using local HNSW vector indexes, you may see different optimizer plans for similarity searches.
• Optimizer Plans for IVF Vector Indexes
  Inverted File Flat (IVF) is a form of Neighbor Partition Vector index. It is a partition-based index that achieves search efficiency by narrowing the search area through the use of neighbor partitions or clusters.
• Vector Index Hints
  If the optimizer does not choose your existing index while running your query and you still want that index to be used, you can rewrite your SQL statement to include vector index hints.