Use in-memory data sets as a base for multiple data models
Duplicating data on in-memory storage is expensive (cost, load times,..). Allowing for the flexibility of having in-memory datasets linked to multiple data models allow for greater flexibility towards end-users while keeping infrastructure price at par. It would also allow for a broader set of customized/tailored data models (mostly related to links between data sets) while still centrally managed.
We are looking into it …
Steve Coleman commented
We're developing our strategy for Azure Analysis Services now, and we're leaning toward having separate models based on how much harmony exists between dims and facts. Sure, we can build a really big model that has all dims and facts together and use perspectives, but then we're committed to a big SKU for AAS to host it on.
We're running into the situation of the same data existing needing to exist in multiple models, so this would be a huge benefit to us to be able to "link" to tables in another model on the instance.
Dylan Morgan commented
How would this work when the metadata is somewhat different? I am curious because this concept would help with semver deployments, but Tabular drops a tables data when the metadata changes (i.e. A column is deleted). I am curious how you would solve for this if the cache is serving multiple metadata stores?
Do perspectives not work for your use case?