DiskANN is a disk-based approximate nearest neighbor search technology specifically designed for processing ultra-large-scale datasets. It uses the Vamana graph algorithm to store data on disk while maintaining efficient vector indexing and retrieval under limited memory conditions.
DiskANN index build parameters
Parameter | Type | Description |
proxima.diskann.builder.pq_dimension_scale | Int | The scale for PQ quantization. PQ dimension = vector dimension/pq_dimension_scale. Default value: 8. |
proxima.diskann.builder.max_degree | Int | The maximum number of output nodes for a node in the DiskANN graph. A larger value results in a more accurate graph but increases the graph construction time. |
proxima.diskann.builder.list_size | Int | The size of the candidate set for edge selection during DiskANN graph construction. A larger value results in a more accurate graph but increases the graph construction time. |
proxima.diskann.builder.thread_count | Int | The number of threads for index building. |
DiskANN index search parameters
Parameter | Type | Description |
proxima.diskann.searcher.list_size | Int | The size of the result candidate set. A larger value results in higher recall but increases query time. |
proxima.diskann.searcher.io_limit | Int | The disk I/O limit for a single query. A maximum of io_limit disk read operations will be performed. This mainly affects the number of Vamana graph walks. A larger value results in higher recall and more I/O operations, which increases query time. |
proxima.diskann.searcher.beam_search_width | Int | The number of parallel I/O operations. |