OpenSearch and the Path to a Trillion-s...
summary
OpenSearch presents concrete engineering changes to reach trillion-scale vectors: 3x search throughput, 3x better disk storage utilization, and 10x faster index builds with Nvidia GPUs (transcript cites GPU index speedups up to ~12x and CPU cost reductions). It adds 32x binary quantization plus disk-backed vectors with reranking to preserve accuracy, supports hybrid text+vector search (8–12% NDCG10 gain), offloads vectors to remote object stores, and provides ML connectors, AI flows and an agent framework for easy onboarding.
AI-written summary · about
chapters (21)
- 0:00OpenSearch introduction
- 0:22Agenda and overview
- 1:47What is a vector in AI
- 3:10Vector similarity search
- 4:37Indexing and searching vectors
- 6:24Trillion-scale vector DB
- 8:11Early use cases and HNSW
- 10:17Semantic search with NLP
- 11:22Scaling with quantization
- 13:08Hybrid text and vector search
- 15:38RAG use cases and workloads
- 17:0232x compression and disk search
- 18:26Hardware acceleration with SIMD/GPU
- 19:09AI agents and trillion-scale
- 20:52Future roadmap for GPU/disk
- 21:53Optimizing disk seeks and routing
- 22:36Multi-tenancy and relevancy
- 23:17Easy vector DB for AI
- 23:59ML connectors and AI flows
- 25:02Agentic framework and MCP
- 25:44Conclusion
related talks
43mOpenSearch and the Path to a Trillion-scale Vector Database
48mUnlocking Next-gen Semantic Search With OpenSearch 3.x
55mVector Search Made Simple: Getting Started with OpenSearch for AI Applications
30mVector Search Made Simple: Getting Started with OpenSearch for AI Applications
05mKeynote: OpenSearch 3.0 and Lucene 10, Paving the Way for AI and Vector Innovation
32mRetrieval Augmentation and Semantic Search at Scale
40mVector Search Made Simple: Getting Started with OpenSearch for AI Applications
47mWhy, What, When and How of Running Diff Types of Sear...