Expand description
DiskANN - Disk-based Approximate Nearest Neighbor Search
Implementation of Microsoft’s DiskANN algorithm for billion-scale vector search. DiskANN is optimized for SSD storage and can handle datasets too large for RAM.
§Key Features
- Disk-based storage with memory-mapped files
- SSD-optimized I/O patterns
- Vamana graph for efficient navigation
- Streaming search without full index loading
- Incremental updates support
§Architecture
- In-memory layer: Hot vectors and navigation graph
- Disk layer: Full vector dataset on SSD
- Compressed layer: Optionally compressed vectors (PQ)
§References
Re-exports§
pub use builder::DiskAnnBuildStats;pub use builder::DiskAnnBuilder;pub use config::DiskAnnConfig;pub use config::PruningStrategy;pub use config::SearchMode;pub use graph::VamanaGraph;pub use graph::VamanaNode;pub use index::DiskAnnIndex;pub use index::IndexMetadata;pub use search::BeamSearch;pub use search::SearchStats;pub use storage::DiskStorage;pub use storage::MemoryMappedStorage;pub use storage::StorageBackend;pub use types::DiskAnnError;pub use types::DiskAnnResult;pub use types::NodeId;pub use types::VectorId;