Petabyte-Scale AI Memory with Serverless Java
summary
Pure Java petabyte AI memory built: Eclipse Store serializes and persists Java object graphs, JVector provides embedded vector search, and Eclipse Data Grid replicates graphs across JVMs with Kafka transport and single-writer eventual consistency. Cell groups run on Kubernetes using GraalVM native images to cut JVM memory by ~100x and start in milliseconds. Claims: real-time access to petabyte knowledge, up to 80% infrastructure cost savings and up to 90% RAM reduction versus traditional caches, proven in Dell integrations.
AI-written summary · about
chapters (20)
- 0:00Introduction to talk
- 1:51Java for AI applications
- 3:42Java object graphs explained
- 6:14Eclipse Serializer project
- 7:39Eclipse Store storage engine
- 9:26Single-node monolith benefits
- 10:08Distributed Java with Data Grid
- 12:57Monolith vs distributed trade-offs
- 15:28Unifying infrastructure layers
- 21:11Java vs SQL for complex queries
- 25:56AI energy crisis challenges
- 27:00Short-term vs long-term AI memory
- 31:05Human brain vs AI architecture
- 34:19Monolithic database inefficiencies
- 39:19Serverless Lambda inspiration
- 41:48Lambda limitations for AI
- 43:14Cell-based Java architecture
- 45:47Peta-scale AI with serverless Java
- 47:17Enterprise AI stack deployment
- 48:47Q&A session
related talks
53mDatabaseless Data Processing - High-Performance for Cloud-Native Apps and AI
44mJava for AI
02h 35mBringing GenAI to the Modern Enterprise. A production use-case. In Serverless Java!
49mThe Java Cloud-Native Stack for Microservices and Serverless Architecture
01h 02mEnabling High Throughput, Low Latency Inference for Your AI Applications
55mLarge Scale Distributed AI on the JVM
52mBuild Smarter Knowledge Management Systems with AI Technology
43mWhy Java and the JVM Will Dominate the Future of Machine Learning, AI, and Big Data