SourceHow AlloyDB ScaNN scales vector search to 10 billion vectors
Organizations building large-scale AI applications can now process massive datasets more efficiently, ensuring faster and more accurate search results for end users.
- Introduced a four-level tree index architecture in preview for AlloyDB ScaNN
- Reduces query traversal complexity to O(N1/4) via hierarchical partitioning
- Enables vector search scaling to over 10 billion vectors
- Achieved <= 51 ms p95 latency and 95% recall in internal tests at 10 billion vectors