FWIW, a primary reason for me to split databases, is nothing to do with
performance, whether it works, (all of which are probably true, hard to prove,
and I accept as true...)
It's because, I can't replace the program with an updated one if live data is
also mingled in with it!
This is probably why you get some people trying to modify programs "on-line"
(which with my capacity for mistakes would be disastrous!)
I inherited one site with mixed program/data. I had to test what I would do at
my site, then go to their site, get everyone out, and hopefully remember the
mod I had tested.
-----
For this reason, I split even a single-user database. OTOH I have had multiple
users share a networked Front-End (A97) with no dire problems, but whether
that is recommended or not is a separate issue from splitting! (I don't
recommend such, I just used a guinea pig customer to try it)(not theory,
actual tests!!!)
-----
I note that Ed, in his first brief reply, correctly identified relevant
issues. Then it got sidetracked by Tony Toews mentioning a potential
performance downside of splitting. Then it got further sidetracked as to
whether "rarely necessary" was an appropriate wording (probably not because it
implied "advisable"). Ed has made it abundantly clear, I think, what will or
may work vs what is recommended best practise.
I have indicated above, that even if unsplit databases work, they are clearly
inconvenient. And since A2000, modifying live programs has become (either
dangerous or impossible), whether or not the program is mixed up with data or
just a split shared Front-End.
No complaints from me, Ed!
Chris