Case Study: DataStax achieves billion-scale vector search performance with OpenSearch and JVector

DataStax scales to 1B vectors with OpenSearch and cuts query latency

DataStax, a data platforms and vector search company, needed a production-ready vector search implementation that could scale to billions of vectors without sacrificing recall or requiring high-RAM, GPU-only setups. They partnered with the vendor OpenSearch to build a solution using a custom plugin for dense vector retrieval, aiming to overcome the limitations of off-the-shelf vector libraries.

Using OpenSearch as a foundation, DataStax implemented their JVector library, which utilized inline vector storage, selective quantization, and concurrent graph construction. This solution delivered key outcomes for DataStax, including lower query latency, less disk I/O, improved index build speed, and reduced infrastructure costs, all while preserving search recall and enabling them to operate effectively at a massive scale.

DataStax