Vector Search Just Got Faster

Dragonfly v1.37 significantly enhances vector search performance by implementing a single global Hierarchical Navigable Small World (HNSW) index, replacing the previous per-shard indexing approach. This update results in up to 7x throughput gains and 65x lower latency compared to earlier versions, while also nearly halving memory usage. The new architecture eliminates instability in precision performance, ensuring deterministic results irrespective of thread count, and offers a tunable balance between throughput and precision. Benchmarks show Dragonfly v1.37 surpasses both its predecessor and competitor Valkey in terms of queries per second and latency, making it a cost-effective solution for production-scale vector workloads. Additionally, v1.37 provides memory improvements for JSON and hash map documents, supports new commands, and includes various bug fixes, marking a substantial advancement in in-memory data store capabilities.