On a extremely big picture 50000 foot level, there's the windows model where you put everything into one piece of software and you run that one piece of software on one server (real or virtual). Its philosophically anti-windows to have more than one piece of software installed per OS. So you want a "something" and a database server, well now you need two installs / licenses / hardware / images / admins / whatever not just one. And in windows philosophy you never run user stuff on a server anyway so right there you may have issues with a "user" app and a "server" db system requiring two systems.
Its hardly all peaches and unicorns on the unix side either. Seriously, you're going to install 17 completely separate, different versions of "magic embedded mysql" or whatever, and have no way to do joins across DB servers or system wide DB backup system? Crazy talk. Why not have every programmer write their own homemade strcmp() complete with different buffer overflow bugs on each version... So the whole concept of a "private" database violates quite a few unix philosophy principles. Handling multiple versions of DBs across different distros is a puzzler for the devs rather than just including one version and calling it good.
The whole mess from the 100K standpoint is DBs and DB apps are too new, too much technical churn, too much incompatibility. For example, why are you installing a new DB and replacing all your API code if all you want is a different engine perhaps (or perhaps not) better suited to your workload? Its conceptually like switching your whole infrastructure to a new word processor in 1985.
Its hardly all peaches and unicorns on the unix side either. Seriously, you're going to install 17 completely separate, different versions of "magic embedded mysql" or whatever, and have no way to do joins across DB servers or system wide DB backup system? Crazy talk. Why not have every programmer write their own homemade strcmp() complete with different buffer overflow bugs on each version... So the whole concept of a "private" database violates quite a few unix philosophy principles. Handling multiple versions of DBs across different distros is a puzzler for the devs rather than just including one version and calling it good.
The whole mess from the 100K standpoint is DBs and DB apps are too new, too much technical churn, too much incompatibility. For example, why are you installing a new DB and replacing all your API code if all you want is a different engine perhaps (or perhaps not) better suited to your workload? Its conceptually like switching your whole infrastructure to a new word processor in 1985.