Every time I read assertions along this line, I’m struck by the ignorance. This advocates for the Exact Same Paradigm as COBOL/VSAM/3270 of 1970: client bound, block mode interface, to passive data. Dr. Codd showed you a better way. The wire is longer, but that doesn’t mean you should cede data integrity to the client; your grandfather coded that way. Send the minimal data to the client, get it back, but don’t trust data integrity to client code. That’s the database’s job. If you want, you can replicate some edits on the client from the catalog constraints. Just don’t trust the client. Remember: the purpose of the relational database is a datastore which is both ignorant of, and agnostic to, any particular application; the datastore accepts requests from ANY application interested in its data. Those that advocate “application databases” are just wrong.

The final nail in this coffin is the growth of slower (at the core/thread level) multi-core processors, which will make the client code slower, not faster. Client application code is inherently linear (won’t benefit from multi-cores; see Amdahl’s Law http://en.wikipedia.org/wiki/Amdahl%27s_law ), while relational database engines are inherently parallel.

Merci pour cet article intéressant. Bien à vous…….