> "I think this point is very important even in the RDBMS side."
This is true. After all you can't extract an audit trail from a deleted record. Simply in terms of information management it often makes sense to represent this as separate info.
> "There are cases, even with relational datastores that would preform better if the dataset was built to the query."
The big problem with doing it that way is that you are screwed as soon as requirements change. I recently blogged about not-1NF designs in PostgreSQL (nested data structures for subset constraints), and the lesson I took away was that you really don't want to have your select queries hitting the same tables you maintain for inserts. You want your queries hitting a normalized data structure even if the data comes in as something different.
This is true. After all you can't extract an audit trail from a deleted record. Simply in terms of information management it often makes sense to represent this as separate info.
> "There are cases, even with relational datastores that would preform better if the dataset was built to the query."
The big problem with doing it that way is that you are screwed as soon as requirements change. I recently blogged about not-1NF designs in PostgreSQL (nested data structures for subset constraints), and the lesson I took away was that you really don't want to have your select queries hitting the same tables you maintain for inserts. You want your queries hitting a normalized data structure even if the data comes in as something different.