The Evolution of Milvus: A Cloud-Native Vector Database
summary
Milvus 2.0 claims a production cloud-native vector database: a fully distributed, horizontally scalable system (v2.2) that disaggregates query, index, and data planes with one coordinator per service and stateless worker clusters, storing logs and index snapshots in S3/MinIO. It uses CPU SIMD (SSE/AVX/AVX512), GPUs/FPGAs and supports Faiss/Annoy-style indexes; a VLDB 2022 paper describes the design. Real deployments include Cleveland Museum art‑lens AI launched in a week and Trend Micro APK threat detection—both working at scale for over 1000 organizations with log-as-data disaggregation and Kubernetes/Helm deployable architecture.
AI-written summary · about
chapters (21)
- 0:00Intro
- 1:15What is Unstructured Data?
- 2:25The Evolution of Data
- 4:29Embed the World
- 6:00Embeddings are Powerful
- 9:03What is a Vector Database?
- 9:26Vector Databases in Production
- 10:20Milvus Features
- 12:34Cloud Native
- 13:27Vector Data ETL
- 15:03Administrative and Visualization Tools
- 15:31Milvus General Architecture
- 16:54Access Layer
- 18:09Coordinator Layer
- 20:05Worker Layer
- 21:32Storage Layer
- 22:28Key Takeaways
- 23:59Art Gallery Experience
- 25:51Threat Detection
- 26:45Property Search
- 27:38Product Recommendation
related talks
40mEvolution of Milvus Cloud-Scalable Vector Database
31mMilvus 2.0: A Vector Database With Cloud-Native Architecture
37mMilvus: Scaling Vector Data Solutions with Gen AI
51mUnleashing the Potential of Cloud-Native Vector Databases
36mMilvus: Scaling Vector Data Solutions with Gen AI
30mMilvus Vector Database: Unlocking the Future of Open Source Vector Databases
35mMilvus 2.0: Building the World's Most Advanced Open-source Vector Database- 37mMilvus: Scaling Vector Data Solutions with Gen AI