This comprehensive study tests the capabilities of Power BI’s new Direct Lake connection to Microsoft’s OneLake platform against three different database sizes, ranging from 100 gigabytes to 10 terabytes, under light to heavy user concurrency. As part of the analysis, analysts compare the test results of Power BI/Direct Lake against an alternative approach, AtScale on Snowflake.
The benchmarking study documents the following key insights:
- Direct Lake serves as a “lazy load” Import Mode alternative with all the same drawbacks and limitations as Import Mode
- Direct Lake is fast on very small data but stumbles with larger data and higher user concurrency
- Upon a data or model refresh, Direct Lake creates a substantial “cold cache” performance hit
- Power BI Web is missing critical modeling features and is poorly suited for multi-user collaboration
- Fabric Lakehouse is not suitable for serving as a general purpose Lakehouse due to its lack of basic data management features
Download the full report for an in-depth explanation of the methodology and results.
