Amazon OpenSearch’s vector engine can now run modern search applications at a third of the cost on OpenSearch 2.17 domains. When you configure a k-NN (vector) index for disk mode, it becomes optimized for operating in a low memory environment. With disk mode on, the index is compressed using techniques like binary quantization and search quality (recall) is retained through a disk-optimized rescoring mechanism using full-precision vectors. Disk-mode is an excellent option for vector search workloads that require high accuracy, cost efficiency and are satisfied by low hundreds-of-milliseconds latency. It provides customers with a lower cost alternative to the existing in-memory mode when single-digit latency is unnecessary. To learn more, refer to the documentation.
Source:: Amazon AWS