The Debate Returns (with more vectors): Which Search Engine?
summary
Five open-source engines compare concrete tradeoffs: one reports 48 billion vectors indexed in a single cluster and adds GPU-accelerated embedding inference for on-node ranking; another focuses on long-lived Lucene-backed stability with vector and learning-to-rank hooks; a vector-first engine emphasizes two-line model integrations and multi-tenancy for millions of tenants; a minimalist engine targets billion-scale similarity retrieval with tight memory/performance tweaks like PQ compression; managed/serverless and S3-backed scaling also appear as operational options.
AI-written summary · about
chapters (10)
- 0:00Introduction and panel setup
- 1:25Panelist introductions
- 8:27Scaling capabilities
- 14:48Unsuitable use cases
- 19:26AI-powered applications
- 27:42Community and open source
- 34:42Choosing another engine
- 40:44Future developments
- 48:58Favorite implementations
- 52:55Closing remarks
related talks
30mVector Search Made Simple: Getting Started with OpenSearch for AI Applications
24mVector Search at Uber
44mVector Search Isn’t Enough
44mVector Search Beyond the Hype: Real-World Implementation and Performance Insights
01h 09mVector Search: Ask Me Anything!
33mBeyond Benchmarks: Real-World AI Search Performance in OpenSearch (Lucene Vs Faiss Vs JVector)
47mWhy, What, When and How of Running Diff Types of Sear...
15mUnconference: Explainable Vector Search