Star Schema ALL the things! But why?
summary
A practical comparison uses NYC bike CSVs (≈212 million rows) loaded into Fabric, measuring refresh time, model size, CPU and query cost. A single flat table is ~6.2 GB in the lakehouse and flat table 212M rows → 6.2GB model, with incremental refresh failing under P1 and full refresh timing out; the star schema compresses to ~3.1 GB and completes a full refresh in ~1h10m, uses ~½ the CPU and runs queries far cheaper while avoiding autoexists correctness traps. Tools: Fabric pipelines, Power BI import mode, Profiler, DAX Studio, Vertipaq analyzer.
AI-written summary · about
chapters (15)
- 0:00Introduction
- 3:40Star schemas in data modeling
- 10:07Flat tables vs star schemas
- 13:20Refresh time and performance
- 17:57Refresh time results
- 21:09Model size and compression
- 25:46Keys and surrogate IDs
- 28:17Reusability and usability
- 31:25Performance and DAX complexity
- 37:27Cost of model choice
- 40:16Query complexity and accuracy
- 43:34Auto exists and results
- 45:16Galaxies and multi-fact models
- 47:01When to adopt star schemas
- 47:43Q&A
related talks
41mThe Unified Star Schema
34mDimensional Data Modeling in the Modern Era
14mBuild your data lakehouse with Starburst Galaxy
26mFrom datasets to tables in a multitenant data lake
41mReal-World NoSQL Schema Design #bbuzz
40mFast Access To Your Complex Data #bbuzz
39mData Warehouses and Multi-Dimensional Data Analysis
33mAnalyzing 33 million bike trips with BigQuery