[ludston]

Good post. I think it is also worth mentioning that the relational model has a few intrinsic performance issues and gotcha's that require some uncomfortable and creative wrangling.

Specifically, safely handling concurrent writes to different rows that are related to each other requires careful consideration, and when you have a join with clauses on both tables it is not possible to have perfect efficiency without multi-table indexes.

I think that these are not intuitive, but solvable. The troubles arise when an application doesn't consider these issues until after it hits scale (which there is sufficient concurrency and enough millions of rows for these problems to rear their heads).