Performance Tuning Apache Solr for Dense Vectors
summary
A 169M-document index of 1,024‑dim vectors starts at ~900–1,000 ms end‑to‑end, 10–11 CPU cores and caps near 50 QPS. Normalizing vectors and switching to dot product cuts CPU to ~4 cores and latency to ~700 ms. An aggressive merge policy reduces segments ~500→200 and drops CPU to ~3 cores and query time substantially. Enabling Java 21 Project Panama pushes CPU under 2 cores. Making vector fields non‑stored shrinks index 1.8 TB→~800 GB and final results reach ~100 ms E2E, <100 ms Q time and >200 QPS.
AI-written summary · about
chapters (12)
- 0:00Introduction
- 0:54Performance isnt free
- 4:00The vision
- 4:58Dense Vectors
- 6:45Performance Benchmarks
- 9:22Horizontal Scaling
- 11:48Merging Segments
- 14:10Project Panama
- 15:25Memory
- 17:09Results
- 18:21Tips
- 18:57Quantization
related talks
41mApache Solr 10: What's Coming up for Vector Search
01h 09mVector Search: Ask Me Anything!
33mBeyond Benchmarks: Real-World AI Search Performance in OpenSearch (Lucene Vs Faiss Vs JVector)
47mLatest innovations from Elasticsearch: Drive speed, scale & relevance (PRT266)
41mPractical Vector Search Configuration, Tuning, RAG, and the Road Ahead
36mSearch Complete
05mKeynote: OpenSearch 3.0 and Lucene 10, Paving the Way for AI and Vector Innovation
32mRetrieval Augmentation and Semantic Search at Scale