The DB is rarely our bottleneck. Our reads are very simple, mostly paged sets from a table with a 1 level join. When we need complex abstractions or transforma

The DB is rarely our bottleneck. Our reads are very simple, mostly paged sets from a table with a 1 level join. When we need complex abstractions or transformations, instead of using function or views, we listen to table changes, and with a TS transformation layer we save a new row in a specific table for the abstraction. That way the BL is handled in TS, and performed in writes asynchronously instead of in-line in reads. The only downside is that the transformed data is not real time. That mechanism allowed us to scale our data very easily without any performance hits.