Keynote | Data Warehouse, Data Lakes, What's Next?
summary
Proposal: a unified HSAP storage called Hologres stores row and column formats, exposes Postgres-compatible SQL, and couples with Flink for realtime dimension joins and feature sinks. Built as a C++ query engine with compute/storage desegregation, it claims 130 million events/sec ingestion peak and 145 million queries/sec mixed serving+analytics at Double 11, and reports 57x YCSB vs HBase on point lookups. The system supports millisecond point lookups, second‑level interactive analytics, and workload isolation.
AI-written summary · about
chapters (20)
- 0:00Introduction
- 1:15Feature Engineering
- 3:30Derived Features
- 4:38Data Patterns
- 5:30Lambda Architecture
- 6:58Storage
- 9:08Hybrid Serving Analytics Processing
- 11:42Data Lakes
- 14:33Data Analysis
- 17:00Holographs
- 20:54Performance
- 21:34Isolation
- 22:32Architecture
- 29:01Structured Tablets
- 30:10Query Plan
- 31:17Nonleaflet Operators
- 31:55Scale
- 33:36Execution Context
- 34:22Cooperative Scheduling
- 37:12Summary
related talks
36mData Warehouse, Data Lakes, What's Next?
28mBreakout Session
25mBuild open format data lakehouse
31mPanel: Open Data Lakes
27mBuilding a PostgreSQL data warehouse
34mCatalogs, Iceberg, and Delta: Building the Foundation for Lakehouse Analytics
40mUnbundling of the Cloud Data Warehouse: Open Source Databases and Data Lakes
31mBuilding large scale, transactional data lakes with Apache Hudi