Paul Turley lays out the significance of composite models in Power BI:
There are been many attempts by Microsoft and other vendors to create a data modelling architecture that provides for fast access to cached data, direct access to live data and scaled-out connections to established data models. Remember ROLAP and HOLAP storage in multidimensional cubes? These were great concepts with significant trade-off limitations. No other vendor has anything like this. Way back in the day, Microsoft jumped on the Ralph Kimball bandwagon to promote the idea that a company should have a “one version of the truth” exposed through their data warehouse and cubes or semantic data models. They met customer demand and gave us a BI tool that, in order to bring data together from multiple sources, makes it easy to create a lot of data silos. Arguably, there are design patterns to minimize data duplication but to use governed datasets, self-service report designers are limited to connecting to large, central models that might only be authored and managed by IT. This new feature can restore balance to the force and bring us back to “one version of the truth” again.
Read on for Paul’s early thoughts on the feature.